1
1
openmpi/ompi/mca/btl/openib/btl_openib_endpoint.c

1531 строка
55 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
* Copyright (c) 2006-2007 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2006-2007 Los Alamos National Security, LLC. All rights
* reserved.
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include <sys/time.h>
#include <time.h>
#include "ompi/types.h"
#include "ompi/mca/pml/base/pml_base_sendreq.h"
#include "orte/mca/ns/base/base.h"
#include "orte/mca/oob/base/base.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/dss/dss.h"
#include "btl_openib.h"
#include "btl_openib_endpoint.h"
#include "btl_openib_proc.h"
#include "btl_openib_frag.h"
#include "ompi/class/ompi_free_list.h"
#include <errno.h>
#include <string.h>
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
#define ENDPOINT_CONNECT_REQ 0
#define ENDPOINT_CONNECT_RSP 1
#define ENDPOINT_CONNECT_ACK 2
static void mca_btl_openib_endpoint_construct(mca_btl_base_endpoint_t* endpoint);
static void mca_btl_openib_endpoint_destruct(mca_btl_base_endpoint_t* endpoint);
int mca_btl_openib_endpoint_create_qp(
mca_btl_openib_module_t* openib_btl,
struct ibv_pd* pd,
struct ibv_cq* cq,
struct ibv_srq* srq,
struct ibv_qp_attr* qp_attr,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
struct ibv_qp** qp,
int qp_idx
);
int mca_btl_openib_endpoint_qp_init_query(
mca_btl_openib_module_t* openib_btl,
struct ibv_qp* qp,
struct ibv_qp_attr* attr,
uint32_t lcl_psn,
uint32_t rem_qp_num,
uint32_t rem_psn,
uint16_t rem_lid,
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
uint32_t rem_mtu,
uint32_t port_num
);
static int post_send(mca_btl_openib_module_t *openib_btl,
mca_btl_openib_endpoint_t *endpoint, mca_btl_openib_frag_t *frag,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
const int qp, const int do_rdma)
{
struct ibv_send_wr *bad_wr;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
assert(!do_rdma || BTL_OPENIB_EAGER_RDMA_QP(qp));
frag->sg_entry.length = frag->segment.seg_len +
sizeof(mca_btl_openib_header_t) +
(do_rdma ? sizeof(mca_btl_openib_footer_t) : 0);
if(frag->sg_entry.length <= openib_btl->ib_inline_max) {
frag->wr_desc.sr_desc.send_flags = IBV_SEND_SIGNALED|IBV_SEND_INLINE;
} else {
frag->wr_desc.sr_desc.send_flags = IBV_SEND_SIGNALED;
}
if(endpoint->nbo)
BTL_OPENIB_HEADER_HTON((*(frag->hdr)));
if(do_rdma) {
mca_btl_openib_footer_t* ftr =
(mca_btl_openib_footer_t*)(((char*)frag->segment.seg_addr.pval) +
frag->segment.seg_len);
frag->wr_desc.sr_desc.opcode = IBV_WR_RDMA_WRITE;
MCA_BTL_OPENIB_RDMA_FRAG_SET_SIZE(ftr, frag->sg_entry.length);
MCA_BTL_OPENIB_RDMA_MAKE_LOCAL(ftr);
#if OMPI_ENABLE_DEBUG
((mca_btl_openib_footer_t*)(((char*)frag->segment.seg_addr.pval) +
frag->segment.seg_len))->seq =
endpoint->eager_rdma_remote.seq++;
#endif
if(endpoint->nbo)
BTL_OPENIB_FOOTER_HTON((*ftr));
frag->wr_desc.sr_desc.wr.rdma.rkey = endpoint->eager_rdma_remote.rkey;
frag->wr_desc.sr_desc.wr.rdma.remote_addr =
endpoint->eager_rdma_remote.base.lval +
endpoint->eager_rdma_remote.head *
openib_btl->eager_rdma_frag_size +
sizeof(mca_btl_openib_header_t) +
mca_btl_openib_component.eager_limit +
sizeof(mca_btl_openib_footer_t);
frag->wr_desc.sr_desc.wr.rdma.remote_addr -= frag->sg_entry.length;
MCA_BTL_OPENIB_RDMA_NEXT_INDEX(endpoint->eager_rdma_remote.head);
} else {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(MCA_BTL_OPENIB_SRQ_QP == endpoint->qps[qp].qp_type) {
frag->wr_desc.sr_desc.opcode = IBV_WR_SEND_WITH_IMM;
frag->wr_desc.sr_desc.imm_data = endpoint->rem_info.rem_index;
} else {
frag->wr_desc.sr_desc.opcode = IBV_WR_SEND;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
frag->base.order = qp;
return ibv_post_send(endpoint->qps[qp].lcl_qp, &frag->wr_desc.sr_desc, &bad_wr);
}
/*
* post a send to the work queue
*/
static int btl_openib_acquire_send_resources(
mca_btl_openib_module_t *openib_btl,
mca_btl_openib_endpoint_t *endpoint,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
mca_btl_openib_frag_t *frag, int *qp, int *do_rdma)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(*do_rdma) {
if(OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
*do_rdma = 0;
} else {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
*qp = mca_btl_openib_component.eager_rdma_qp;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(OPAL_THREAD_ADD32(&endpoint->qps[*qp].sd_wqe, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->qps[*qp].sd_wqe, 1);
if(*do_rdma)
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
opal_list_append(&endpoint->qps[*qp].pending_frags,
(opal_list_item_t *)frag);
return OMPI_ERR_OUT_OF_RESOURCE;
}
if(*do_rdma)
return OMPI_SUCCESS;
if(MCA_BTL_OPENIB_PP_QP == endpoint->qps[*qp].qp_type) {
if(OPAL_THREAD_ADD32(&endpoint->qps[*qp].u.pp_qp.sd_credits, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->qps[*qp].u.pp_qp.sd_credits, 1);
OPAL_THREAD_ADD32(&endpoint->qps[*qp].sd_wqe, 1);
opal_list_append(&endpoint->qps[*qp].pending_frags,
(opal_list_item_t *)frag);
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(OPAL_THREAD_ADD32(&openib_btl->qps[*qp].u.srq_qp.sd_credits, -1) < 0) {
OPAL_THREAD_ADD32(&openib_btl->qps[*qp].u.srq_qp.sd_credits, 1);
OPAL_THREAD_ADD32(&endpoint->qps[*qp].sd_wqe, 1);
OPAL_THREAD_LOCK(&openib_btl->ib_lock);
opal_list_append(&openib_btl->qps[*qp].u.srq_qp.pending_frags,
(opal_list_item_t *)frag);
OPAL_THREAD_UNLOCK(&openib_btl->ib_lock);
return OMPI_ERR_OUT_OF_RESOURCE;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
}
return OMPI_SUCCESS;
}
#define GET_CREDITS(FROM, TO) \
do { \
TO = FROM; \
} while(0 == OPAL_ATOMIC_CMPSET_32(&FROM, TO, 0))
/* this function os called with endpoint->endpoint_lock held */
static inline int mca_btl_openib_endpoint_post_send(mca_btl_openib_module_t* openib_btl,
mca_btl_openib_endpoint_t * endpoint,
mca_btl_openib_frag_t * frag)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int do_rdma = 0, qp, ib_rc;
int32_t cm_return;
frag->sg_entry.addr = (unsigned long) frag->hdr;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(frag->base.order != MCA_BTL_NO_ORDER) {
qp = frag->base.order; /* if order is provided use it */
} else {
qp = frag->qp_idx;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(frag->segment.seg_len <= mca_btl_openib_component.eager_limit &&
(frag->base.des_flags & MCA_BTL_DES_FLAGS_PRIORITY))
do_rdma = 1; /* High priority frag. Try to send over eager RDMA */
}
if(btl_openib_acquire_send_resources(openib_btl, endpoint, frag, &qp,
&do_rdma) == OMPI_ERR_OUT_OF_RESOURCE)
return OMPI_SUCCESS;
frag->hdr->credits = 0;
if(BTL_OPENIB_EAGER_RDMA_QP(qp)) {
GET_CREDITS(endpoint->eager_rdma_local.credits, frag->hdr->credits);
if(frag->hdr->credits)
frag->hdr->credits |= BTL_OPENIB_RDMA_CREDITS_FLAG;
}
if(MCA_BTL_OPENIB_PP_QP == endpoint->qps[qp].qp_type &&
0 == frag->hdr->credits) {
GET_CREDITS(endpoint->qps[qp].u.pp_qp.rd_credits, frag->hdr->credits);
}
GET_CREDITS(endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
/* cm_seen is only 8 bytes, but cm_return is 32 bytes */
if(cm_return > 255) {
frag->hdr->cm_seen = 255;
cm_return -= 255;
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
} else {
frag->hdr->cm_seen = cm_return;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
ib_rc = post_send(openib_btl, endpoint, frag, qp, do_rdma);
if(ib_rc) {
if(endpoint->nbo) {
BTL_OPENIB_HEADER_NTOH((*(frag->hdr)));
}
if(BTL_OPENIB_IS_RDMA_CREDITS(frag->hdr->credits)) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_local.credits,
BTL_OPENIB_CREDITS(frag->hdr->credits));
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OPAL_THREAD_ADD32(&endpoint->qps[qp].sd_wqe, 1);
if(do_rdma) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
} else {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(MCA_BTL_OPENIB_PP_QP == endpoint->qps[qp].qp_type) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.rd_credits, frag->hdr->credits);
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, 1);
} else {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, 1);
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_ERROR(("error posting send request error %d: %s\n",
ib_rc, strerror(ib_rc)));
return OMPI_ERROR;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
#if 0
mca_btl_openib_post_srr_all(openib_btl, 1);
mca_btl_openib_endpoint_post_rr_all(endpoint, 1);
#endif
return OMPI_SUCCESS;
}
OBJ_CLASS_INSTANCE(mca_btl_openib_endpoint_t,
opal_list_item_t, mca_btl_openib_endpoint_construct,
mca_btl_openib_endpoint_destruct);
/*
* Initialize state of the endpoint instance.
*
*/
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
static void mca_btl_openib_endpoint_construct_qp(mca_btl_base_endpoint_t *endpoint, int qp)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
bool pp = (MCA_BTL_OPENIB_PP_QP ==
mca_btl_openib_component.qp_infos[qp].type);
endpoint->qps[qp].lcl_qp_attr =
(struct ibv_qp_attr *)malloc(sizeof(struct ibv_qp_attr));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
memset(endpoint->qps[qp].lcl_qp_attr, 0, sizeof(struct ibv_qp_attr));
endpoint->qps[qp].qp_type = mca_btl_openib_component.qp_infos[qp].type;
endpoint->qps[qp].lcl_qp = NULL;
endpoint->qps[qp].rd_pending_credit_chks = 0;
/* setup rem_info */
endpoint->rem_info.rem_qps[qp].rem_qp_num = 0;
endpoint->rem_info.rem_qps[qp].rem_psn = 0;
OBJ_CONSTRUCT(&endpoint->qps[qp].pending_frags, opal_list_t);
if(pp) {
/* local credits are set here such that on initial posting
* of the receive buffers we end up with zero credits to return
* to our peer. The peer initializes his sd_credits to reflect this
* below. Note that this may be a problem for iWARP as the sender
* now has credits even if the receive buffers are not yet posted
*/
endpoint->qps[qp].u.pp_qp.rd_credits =
-mca_btl_openib_component.qp_infos[qp].rd_num;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->qps[qp].u.pp_qp.rd_posted = 0;
endpoint->qps[qp].u.pp_qp.cm_sent = 0;
endpoint->qps[qp].u.pp_qp.cm_return =
-mca_btl_openib_component.qp_infos[qp].u.pp_qp.rd_rsv;
endpoint->qps[qp].u.pp_qp.cm_received =
mca_btl_openib_component.qp_infos[qp].u.pp_qp.rd_rsv;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* initialize the local view of credits */
endpoint->qps[qp].u.pp_qp.sd_credits =
mca_btl_openib_component.qp_infos[qp].rd_num;
/* number of available send wqes */
endpoint->qps[qp].sd_wqe = mca_btl_openib_component.qp_infos[qp].rd_num;
} else {
/* number of available send wqes */
endpoint->qps[qp].sd_wqe = mca_btl_openib_component.qp_infos[qp].u.srq_qp.sd_max;
}
}
static void mca_btl_openib_endpoint_construct(mca_btl_base_endpoint_t* endpoint)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int qp;
/* setup qp structures */
if( mca_btl_openib_component.num_qps > 0 ) {
endpoint->qps =
(mca_btl_openib_endpoint_qp_t*)
malloc(sizeof(mca_btl_openib_endpoint_qp_t) *
mca_btl_openib_component.num_qps);
endpoint->rem_info.rem_qps =
(mca_btl_openib_rem_qp_info_t*)
malloc(sizeof(mca_btl_openib_rem_qp_info_t) *
mca_btl_openib_component.num_qps);
}
endpoint->endpoint_btl = 0;
endpoint->endpoint_proc = 0;
endpoint->endpoint_tstamp = 0.0;
endpoint->endpoint_state = MCA_BTL_IB_CLOSED;
endpoint->endpoint_retries = 0;
OBJ_CONSTRUCT(&endpoint->endpoint_lock, opal_mutex_t);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OBJ_CONSTRUCT(&endpoint->pending_lazy_frags, opal_list_t);
OBJ_CONSTRUCT(&endpoint->pending_get_frags, opal_list_t);
OBJ_CONSTRUCT(&endpoint->pending_put_frags, opal_list_t);
endpoint->get_tokens = mca_btl_openib_component.ib_qp_ous_rd_atom;
/* initialize RDMA eager related parts */
endpoint->eager_recv_count = 0;
memset(&endpoint->eager_rdma_remote, 0,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
sizeof(mca_btl_openib_eager_rdma_remote_t));
memset(&endpoint->eager_rdma_local, 0,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
sizeof(mca_btl_openib_eager_rdma_local_t));
OBJ_CONSTRUCT(&endpoint->eager_rdma_local.lock, opal_mutex_t);
endpoint->rem_info.rem_lid = 0;
endpoint->rem_info.rem_subnet_id = 0;
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
endpoint->rem_info.rem_mtu = 0;
endpoint->nbo = false;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->use_eager_rdma = false;
endpoint->eager_rdma_remote.tokens = 0;
endpoint->eager_rdma_local.credits = 0;
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
mca_btl_openib_endpoint_construct_qp(endpoint, qp);
}
}
/*
* Destroy a endpoint
*
*/
static void mca_btl_openib_endpoint_destruct(mca_btl_base_endpoint_t* endpoint)
{
bool pval_clean = false;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int qp;
/* Release memory resources */
do {
/* Make sure that mca_btl_openib_endpoint_connect_eager_rdma ()
* was not in "connect" or "bad" flow (failed to allocate memory)
* and changed the pointer back to NULL
*/
if(!opal_atomic_cmpset_ptr(&endpoint->eager_rdma_local.base.pval, NULL,
(void*)1)) {
if ((void*)1 != endpoint->eager_rdma_local.base.pval &&
NULL != endpoint->eager_rdma_local.base.pval) {
endpoint->endpoint_btl->super.btl_mpool->mpool_free(endpoint->endpoint_btl->super.btl_mpool,
endpoint->eager_rdma_local.base.pval,
(mca_mpool_base_registration_t*)endpoint->eager_rdma_local.reg);
pval_clean=true;
}
} else {
pval_clean=true;
}
} while (!pval_clean);
/* Close opened QPs if we have them*/
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(MCA_BTL_IB_CLOSED != endpoint->endpoint_state) {
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
OBJ_DESTRUCT(&endpoint->qps[qp].pending_frags);
MCA_BTL_OPENIB_CLEAN_PENDING_FRAGS(endpoint->endpoint_btl,
&endpoint->qps[qp].pending_frags);
if(ibv_destroy_qp(endpoint->qps[qp].lcl_qp)) {
BTL_ERROR(("Failed to destroy QP:%d\n", qp));
}
free(endpoint->qps[qp].lcl_qp_attr);
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* free the qps */
free(endpoint->qps);
}
OBJ_DESTRUCT(&endpoint->endpoint_lock);
/* Clean pending lists */
MCA_BTL_OPENIB_CLEAN_PENDING_FRAGS(endpoint->endpoint_btl,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
&endpoint->pending_lazy_frags);
OBJ_DESTRUCT(&endpoint->pending_lazy_frags);
MCA_BTL_OPENIB_CLEAN_PENDING_FRAGS(endpoint->endpoint_btl,
&endpoint->pending_get_frags);
OBJ_DESTRUCT(&endpoint->pending_get_frags);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
MCA_BTL_OPENIB_CLEAN_PENDING_FRAGS(endpoint->endpoint_btl,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
&endpoint->pending_put_frags);
OBJ_DESTRUCT(&endpoint->pending_put_frags);
}
/*
* Send connection information to remote endpoint using OOB
*
*/
static void mca_btl_openib_endpoint_send_cb(
int status,
orte_process_name_t* endpoint,
orte_buffer_t* buffer,
orte_rml_tag_t tag,
void* cbdata)
{
OBJ_RELEASE(buffer);
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
static int mca_btl_openib_endpoint_send_connect_data(mca_btl_base_endpoint_t* endpoint, uint8_t message_type)
{
orte_buffer_t* buffer = OBJ_NEW(orte_buffer_t);
int rc;
if(NULL == buffer) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* pack the info in the send buffer */
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT8);
rc = orte_dss.pack(buffer, &message_type, 1, ORTE_UINT8);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT64);
rc = orte_dss.pack(buffer, &endpoint->subnet_id, 1, ORTE_UINT64);
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/*GMS left off here */
if(message_type != ENDPOINT_CONNECT_REQ) {
/* send the QP connect request info we respond to */
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT32);
rc = orte_dss.pack(buffer,
&endpoint->rem_info.rem_qps[0].rem_qp_num, 1,
ORTE_UINT32);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT16);
rc = orte_dss.pack(buffer, &endpoint->rem_info.rem_lid, 1, ORTE_UINT16);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(message_type != ENDPOINT_CONNECT_ACK) {
int qp;
/* stuff all the QP info into the buffer */
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT32);
rc = orte_dss.pack(buffer, &endpoint->qps[qp].lcl_qp->qp_num,
1, ORTE_UINT32);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT32);
rc = orte_dss.pack(buffer, &endpoint->qps[qp].lcl_psn, 1,
ORTE_UINT32);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT16);
rc = orte_dss.pack(buffer, &endpoint->endpoint_btl->lid, 1, ORTE_UINT16);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT32);
rc = orte_dss.pack(buffer, &endpoint->endpoint_btl->hca->mtu, 1,
ORTE_UINT32);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_output(mca_btl_base_output, "packing %d of %d\n", 1, ORTE_UINT32);
rc = orte_dss.pack(buffer, &endpoint->index, 1, ORTE_UINT32);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* send to endpoint */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
rc = orte_rml.send_buffer_nb(&endpoint->endpoint_proc->proc_guid,
buffer, ORTE_RML_TAG_DYNAMIC-1, 0,
mca_btl_openib_endpoint_send_cb, NULL);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_VERBOSE(("Sent QP Info, LID = %d, SUBNET = %016x\n",
endpoint->endpoint_btl->lid,
endpoint->subnet_id));
if(rc < 0) {
ORTE_ERROR_LOG(rc);
return rc;
}
return OMPI_SUCCESS;
}
/*
* Set remote connection info
* (from OOB connection)
*
*/
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
static int mca_btl_openib_endpoint_set_remote_info(mca_btl_base_endpoint_t* endpoint,
mca_btl_openib_rem_info_t* rem_info)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* copy the rem_info stuff */
memcpy(&((mca_btl_openib_endpoint_t*) endpoint)->rem_info,
rem_info, sizeof(mca_btl_openib_rem_info_t));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* copy over the rem qp info */
memcpy(endpoint->rem_info.rem_qps,
rem_info->rem_qps, sizeof(mca_btl_openib_rem_qp_info_t) *
mca_btl_openib_component.num_qps);
BTL_VERBOSE(("Setting QP info, LID = %d", endpoint->rem_info.rem_lid));
return ORTE_SUCCESS;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
static int mca_btl_openib_endpoint_qp_create_one(
mca_btl_base_endpoint_t* endpoint, int prio, int qp)
{
int rc;
mca_btl_openib_module_t *openib_btl =
(mca_btl_openib_module_t*)endpoint->endpoint_btl;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
struct ibv_srq *srq =
(MCA_BTL_OPENIB_PP_QP == openib_btl->qps[qp].type) ? NULL :
openib_btl->qps[qp].u.srq_qp.srq;
/* Create the Queue Pair */
if(OMPI_SUCCESS != (rc = mca_btl_openib_endpoint_create_qp(openib_btl,
openib_btl->hca->ib_pd,
openib_btl->ib_cq[prio],
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
srq,
endpoint->qps[qp].lcl_qp_attr,
&endpoint->qps[qp].lcl_qp,
qp))) {
BTL_ERROR(("error creating queue pair, error code %d", rc));
return rc;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->qps[qp].lcl_psn = lrand48() & 0xffffff;
endpoint->qps[qp].credit_frag = NULL;
openib_btl->cq_users[prio]++;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(MCA_BTL_OPENIB_SRQ_QP == endpoint->qps[qp].qp_type) {
mca_btl_openib_post_srr(openib_btl, 1, qp);
} else {
mca_btl_openib_endpoint_post_rr(endpoint, 1, qp);
}
return OMPI_SUCCESS;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
static int mca_btl_openib_endpoint_qp_create_all(
mca_btl_base_endpoint_t* endpoint)
{
int qp, rc;
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
if( mca_btl_openib_component.qp_infos[qp].size <=
mca_btl_openib_component.eager_limit) {
rc = mca_btl_openib_endpoint_qp_create_one(endpoint,
BTL_OPENIB_HP_CQ, qp);
if(rc != OMPI_SUCCESS)
return rc;
} else {
rc = mca_btl_openib_endpoint_qp_create_one(endpoint,
BTL_OPENIB_LP_CQ, qp);
if(rc != OMPI_SUCCESS)
return rc;
}
}
return OMPI_SUCCESS;
}
/*
* Start to connect to the endpoint. We send our Queue Pair
* information over the TCP OOB communication mechanism.
* On completion of our send, a send completion handler
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
* is called
*
*/
static int mca_btl_openib_endpoint_start_connect(mca_btl_base_endpoint_t* endpoint)
{
int rc;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_VERBOSE(("Initialized QP , LID = %d",
((mca_btl_openib_module_t*)endpoint->endpoint_btl)->lid));
rc = mca_btl_openib_endpoint_qp_create_all(endpoint);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(rc != OMPI_SUCCESS)
return rc;
/* Send connection info over to remote endpoint */
endpoint->endpoint_state = MCA_BTL_IB_CONNECTING;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(OMPI_SUCCESS != (rc = mca_btl_openib_endpoint_send_connect_data(endpoint,
ENDPOINT_CONNECT_REQ))) {
BTL_ERROR(("error sending connect request, error code %d", rc));
return rc;
}
return OMPI_SUCCESS;
}
/*
* Reply to a `start - connect' message
*
*/
static int mca_btl_openib_endpoint_reply_start_connect(
mca_btl_openib_endpoint_t *endpoint,
mca_btl_openib_rem_info_t *rem_info)
{
int rc;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_VERBOSE(("Initialized QPs, LID = %d",
((mca_btl_openib_module_t*)endpoint->endpoint_btl)->lid));
rc = mca_btl_openib_endpoint_qp_create_all(endpoint);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(rc != OMPI_SUCCESS)
return rc;
/* Set the remote side info */
mca_btl_openib_endpoint_set_remote_info(endpoint, rem_info);
/* Connect to endpoint */
rc = mca_btl_openib_endpoint_connect(endpoint);
if(rc != OMPI_SUCCESS) {
BTL_ERROR(("error in endpoint connect error code is %d", rc));
return rc;
}
/* Send connection info over to remote endpoint */
endpoint->endpoint_state = MCA_BTL_IB_CONNECT_ACK;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(OMPI_SUCCESS != (rc = mca_btl_openib_endpoint_send_connect_data(endpoint,
ENDPOINT_CONNECT_RSP))) {
BTL_ERROR(("error in endpoint send connect request error code is %d",
rc));
return rc;
}
return OMPI_SUCCESS;
}
/*
* called when the openib has completed setup via the
* OOB channel
*/
static void mca_btl_openib_endpoint_connected(mca_btl_openib_endpoint_t *endpoint)
{
opal_list_item_t *frag_item;
mca_btl_openib_frag_t *frag;
mca_btl_openib_module_t* openib_btl;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->endpoint_state = MCA_BTL_IB_CONNECTED;
endpoint->endpoint_btl->poll_cq = true;
/**
* The connection is correctly setup. Now we can decrease the event trigger.
*/
opal_progress_event_users_decrement();
/* While there are frags in the list,
* process them */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
while(!opal_list_is_empty(&(endpoint->pending_lazy_frags))) {
frag_item = opal_list_remove_first(&(endpoint->pending_lazy_frags));
frag = (mca_btl_openib_frag_t *) frag_item;
openib_btl = endpoint->endpoint_btl;
/* We need to post this one */
if(OMPI_SUCCESS != mca_btl_openib_endpoint_post_send(openib_btl, endpoint, frag))
BTL_ERROR(("Error posting send"));
}
}
/*
* Non blocking OOB recv callback.
* Read incoming QP and other info, and if this endpoint
* is trying to connect, reply with our QP info,
* otherwise try to modify QP's and establish
* reliable connection
*
*/
static void mca_btl_openib_endpoint_recv(
int status,
orte_process_name_t* process_name,
orte_buffer_t* buffer,
orte_rml_tag_t tag,
void* cbdata)
{
mca_btl_openib_proc_t *ib_proc;
mca_btl_openib_endpoint_t *ib_endpoint = NULL;
int endpoint_state;
int rc;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
uint32_t i, lcl_qp = 0;
uint16_t lcl_lid = 0;
int32_t cnt = 1;
mca_btl_openib_rem_info_t rem_info;
uint8_t message_type;
bool master;
/* start by unpacking data first so we know who is knocking at
our door */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT8);
rc = orte_dss.unpack(buffer, &message_type, &cnt, ORTE_UINT8);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT64);
rc = orte_dss.unpack(buffer, &rem_info.rem_subnet_id, &cnt, ORTE_UINT64);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(message_type != ENDPOINT_CONNECT_REQ) {
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT32);
rc = orte_dss.unpack(buffer, &lcl_qp, &cnt, ORTE_UINT32);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT16);
rc = orte_dss.unpack(buffer, &lcl_lid, &cnt, ORTE_UINT16);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(message_type != ENDPOINT_CONNECT_ACK) {
int qp;
/* get ready for the data */
rem_info.rem_qps =
(mca_btl_openib_rem_qp_info_t*) malloc(sizeof(mca_btl_openib_rem_qp_info_t) *
mca_btl_openib_component.num_qps);
/* unpack all the qp info */
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT32);
rc = orte_dss.unpack(buffer, &rem_info.rem_qps[qp].rem_qp_num, &cnt,
ORTE_UINT32);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT32);
rc = orte_dss.unpack(buffer, &rem_info.rem_qps[qp].rem_psn, &cnt,
ORTE_UINT32);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
}
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT16);
rc = orte_dss.unpack(buffer, &rem_info.rem_lid, &cnt, ORTE_UINT16);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT32);
rc = orte_dss.unpack(buffer, &rem_info.rem_mtu, &cnt, ORTE_UINT32);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
opal_output(mca_btl_base_output, "unpacking %d of %d\n", cnt, ORTE_UINT32);
rc = orte_dss.unpack(buffer, &rem_info.rem_index, &cnt, ORTE_UINT32);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_VERBOSE(("Received QP Info, LID = %d, SUBNET = %016x\n",
rem_info.rem_lid,
rem_info.rem_subnet_id));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
master = orte_ns.compare_fields(ORTE_NS_CMP_ALL, orte_process_info.my_name,
process_name) > 0 ? true : false;
for(ib_proc = (mca_btl_openib_proc_t*)
opal_list_get_first(&mca_btl_openib_component.ib_procs);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
ib_proc != (mca_btl_openib_proc_t*)
opal_list_get_end(&mca_btl_openib_component.ib_procs);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
ib_proc = (mca_btl_openib_proc_t*)opal_list_get_next(ib_proc)) {
bool found = false;
if(orte_ns.compare_fields(ORTE_NS_CMP_ALL,
&ib_proc->proc_guid, process_name) != ORTE_EQUAL)
continue;
if(message_type != ENDPOINT_CONNECT_REQ) {
/* This is a reply message. Try to get the endpoint instance the
* reply belongs to */
for(i = 0; i < ib_proc->proc_endpoint_count; i++) {
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
ib_endpoint = ib_proc->proc_endpoints[i];
if(ib_endpoint->qps[0].lcl_qp != NULL &&
lcl_lid == ib_endpoint->endpoint_btl->lid &&
lcl_qp == ib_endpoint->qps[0].lcl_qp->qp_num &&
rem_info.rem_subnet_id == ib_endpoint->subnet_id) {
found = true;
break;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
} else {
/* This is new connection request. If this is master try to find
* endpoint in a connecting state. If this is slave try to find
* endpoint in closed state and initiate connection back */
mca_btl_openib_endpoint_t *ib_endpoint_found = NULL;
for(i = 0; i < ib_proc->proc_endpoint_count; i++) {
ib_endpoint = ib_proc->proc_endpoints[i];
if(ib_endpoint->subnet_id != rem_info.rem_subnet_id ||
(ib_endpoint->endpoint_state != MCA_BTL_IB_CONNECTING
&& ib_endpoint->endpoint_state != MCA_BTL_IB_CLOSED))
continue;
found = true;
ib_endpoint_found = ib_endpoint;
if((master &&
ib_endpoint->endpoint_state == MCA_BTL_IB_CONNECTING) ||
(!master &&
ib_endpoint->endpoint_state == MCA_BTL_IB_CLOSED))
break; /* Found one. No point to continue */
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
ib_endpoint = ib_endpoint_found;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* if this is slave and there is no endpoints in closed state
* then all connection are already in progress so just ignore
* this connection request */
if(found && !master &&
ib_endpoint->endpoint_state != MCA_BTL_IB_CLOSED) {
return;
}
}
if(!found) {
BTL_ERROR(("can't find suitable endpoint for this peer\n"));
return;
}
endpoint_state = ib_endpoint->endpoint_state;
/* Update status */
switch(endpoint_state) {
case MCA_BTL_IB_CLOSED :
/* We had this connection closed before.
* The endpoint is trying to connect. Move the
* status of this connection to CONNECTING,
* and then reply with our QP information */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(master) {
rc = mca_btl_openib_endpoint_reply_start_connect(ib_endpoint, &rem_info);
} else {
rc = mca_btl_openib_endpoint_start_connect(ib_endpoint);
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(OMPI_SUCCESS != rc) {
BTL_ERROR(("error in endpoint reply start connect"));
break;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
}
/** As long as we expect a message from the peer (in order to setup the connection)
* let the event engine pool the OOB events. Note: we increment it once peer active
* connection.
*/
opal_progress_event_users_increment();
break;
case MCA_BTL_IB_CONNECTING :
mca_btl_openib_endpoint_set_remote_info(ib_endpoint, &rem_info);
if(OMPI_SUCCESS != (rc = mca_btl_openib_endpoint_connect(ib_endpoint))) {
BTL_ERROR(("endpoint connect error: %d", rc));
break;
}
if(master) {
ib_endpoint->endpoint_state = MCA_BTL_IB_WAITING_ACK;
/* Send him an ack */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
mca_btl_openib_endpoint_send_connect_data(ib_endpoint,
ENDPOINT_CONNECT_RSP);
} else {
mca_btl_openib_endpoint_send_connect_data(ib_endpoint,
ENDPOINT_CONNECT_ACK);
mca_btl_openib_endpoint_connected(ib_endpoint);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
}
break;
case MCA_BTL_IB_WAITING_ACK:
mca_btl_openib_endpoint_connected(ib_endpoint);
break;
case MCA_BTL_IB_CONNECT_ACK:
mca_btl_openib_endpoint_send_connect_data(ib_endpoint,
ENDPOINT_CONNECT_ACK);
mca_btl_openib_endpoint_connected(ib_endpoint);
break;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
case MCA_BTL_IB_CONNECTED :
break;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
default :
BTL_ERROR(("Invalid endpoint state %d", endpoint_state));
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OPAL_THREAD_UNLOCK(&ib_endpoint->endpoint_lock);
break;
}
}
/*
* Post the OOB recv (for receiving the peers information)
*/
void mca_btl_openib_post_recv()
{
orte_rml.recv_buffer_nb(
ORTE_NAME_WILDCARD,
ORTE_RML_TAG_DYNAMIC-1,
ORTE_RML_PERSISTENT,
mca_btl_openib_endpoint_recv,
NULL);
}
/*
* Attempt to send a fragment using a given endpoint. If the endpoint is not
* connected, queue the fragment and start the connection as required.
*/
int mca_btl_openib_endpoint_send(
mca_btl_base_endpoint_t* endpoint,
mca_btl_openib_frag_t* frag
)
{
int rc;
bool call_progress = false;
mca_btl_openib_module_t *openib_btl;
OPAL_THREAD_LOCK(&endpoint->endpoint_lock);
switch(endpoint->endpoint_state) {
case MCA_BTL_IB_CONNECTING:
BTL_VERBOSE(("Queing because state is connecting"));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_list_append(&endpoint->pending_lazy_frags,
(opal_list_item_t *)frag);
call_progress = true;
rc = OMPI_SUCCESS;
break;
case MCA_BTL_IB_CONNECT_ACK:
case MCA_BTL_IB_WAITING_ACK:
BTL_VERBOSE(("Queuing because waiting for ack"));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_list_append(&endpoint->pending_lazy_frags,
(opal_list_item_t *)frag);
call_progress = true;
rc = OMPI_SUCCESS;
break;
case MCA_BTL_IB_CLOSED:
BTL_VERBOSE(("Connection to endpoint closed ... connecting ..."));
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
opal_list_append(&endpoint->pending_lazy_frags,
(opal_list_item_t *)frag);
rc = mca_btl_openib_endpoint_start_connect(endpoint);
/**
* As long as we expect a message from the peer (in order to setup the connection)
* let the event engine pool the OOB events. Note: we increment it once peer active
* connection.
*/
opal_progress_event_users_increment();
call_progress = true;
break;
case MCA_BTL_IB_FAILED:
rc = OMPI_ERR_UNREACH;
break;
case MCA_BTL_IB_CONNECTED:
{
openib_btl = endpoint->endpoint_btl;
BTL_VERBOSE(("Send to : %d, len : %lu, frag : %p",
endpoint->endpoint_proc->proc_guid.vpid,
frag->sg_entry.length,
frag));
rc = mca_btl_openib_endpoint_post_send(openib_btl, endpoint, frag);
break;
}
default:
rc = OMPI_ERR_UNREACH;
}
OPAL_THREAD_UNLOCK(&endpoint->endpoint_lock);
if(call_progress) opal_progress();
return rc;
}
/*
* Complete connection to endpoint.
*/
static int mca_btl_openib_endpoint_connect_qp(
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
mca_btl_openib_endpoint_t *endpoint, int prio, int qp)
{
int rc;
mca_btl_openib_module_t* openib_btl =
(mca_btl_openib_module_t*)endpoint->endpoint_btl;
/* Connection establishment RC */
rc = mca_btl_openib_endpoint_qp_init_query(openib_btl,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->qps[qp].lcl_qp,
endpoint->qps[qp].lcl_qp_attr,
endpoint->qps[qp].lcl_psn,
endpoint->rem_info.rem_qps[qp].rem_qp_num,
endpoint->rem_info.rem_qps[qp].rem_psn,
endpoint->rem_info.rem_lid,
endpoint->rem_info.rem_mtu,
openib_btl->port_num);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
return rc;
}
int mca_btl_openib_endpoint_connect(
mca_btl_openib_endpoint_t *endpoint)
{
int rc;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int qp;
for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
if( mca_btl_openib_component.qp_infos[qp].size <= mca_btl_openib_component.eager_limit) {
rc = mca_btl_openib_endpoint_connect_qp(endpoint,
BTL_OPENIB_HP_CQ,
qp);
} else {
rc = mca_btl_openib_endpoint_connect_qp(endpoint,
BTL_OPENIB_LP_CQ,
qp);
}
if(rc != OMPI_SUCCESS) {
return rc;
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
return OMPI_SUCCESS;
}
/*
* Create the queue pair note that this is just the initial
* queue pair creation and we need to get the remote queue pair
* info from the peer before the qp is usable,
*/
int mca_btl_openib_endpoint_create_qp(
mca_btl_openib_module_t* openib_btl,
struct ibv_pd* pd,
struct ibv_cq* cq,
struct ibv_srq* srq,
struct ibv_qp_attr* qp_attr,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
struct ibv_qp** qp,
int qp_idx
)
{
{
struct ibv_qp* my_qp;
struct ibv_qp_init_attr qp_init_attr;
memset(&qp_init_attr, 0, sizeof(struct ibv_qp_init_attr));
qp_init_attr.send_cq = cq;
qp_init_attr.recv_cq = cq;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(MCA_BTL_OPENIB_PP_QP == mca_btl_openib_component.qp_infos[qp_idx].type) {
qp_init_attr.cap.max_recv_wr =
mca_btl_openib_component.qp_infos[qp_idx].rd_num +
mca_btl_openib_component.qp_infos[qp_idx].u.pp_qp.rd_rsv;
qp_init_attr.cap.max_send_wr =
mca_btl_openib_component.qp_infos[qp_idx].rd_num + 1;
} else {
qp_init_attr.cap.max_recv_wr =
mca_btl_openib_component.qp_infos[qp_idx].rd_num;
qp_init_attr.cap.max_send_wr =
mca_btl_openib_component.qp_infos[qp_idx].u.srq_qp.sd_max +
BTL_OPENIB_EAGER_RDMA_QP(qp_idx);
}
qp_init_attr.cap.max_send_sge = mca_btl_openib_component.ib_sg_list_size;
qp_init_attr.cap.max_recv_sge = mca_btl_openib_component.ib_sg_list_size;
qp_init_attr.qp_type = IBV_QPT_RC;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
qp_init_attr.srq = srq;
my_qp = ibv_create_qp(pd, &qp_init_attr);
if(NULL == my_qp) {
BTL_ERROR(("error creating qp errno says %s", strerror(errno)));
return OMPI_ERROR;
}
(*qp) = my_qp;
openib_btl->ib_inline_max = qp_init_attr.cap.max_inline_data;
}
{
qp_attr->qp_state = IBV_QPS_INIT;
qp_attr->pkey_index = openib_btl->pkey_index; /*mca_btl_openib_component.ib_pkey_ix; */
qp_attr->port_num = openib_btl->port_num;
qp_attr->qp_access_flags = IBV_ACCESS_REMOTE_WRITE | IBV_ACCESS_REMOTE_READ;
if(ibv_modify_qp((*qp), qp_attr,
IBV_QP_STATE |
IBV_QP_PKEY_INDEX |
IBV_QP_PORT |
IBV_QP_ACCESS_FLAGS )) {
BTL_ERROR(("error modifying qp to INIT errno says %s", strerror(errno)));
return OMPI_ERROR;
}
}
return OMPI_SUCCESS;
}
/*
* The queue pair has been created and we have received the remote
* queue pair information from the peer so we init this queue pair
* and are ready to roll.
*/
int mca_btl_openib_endpoint_qp_init_query(
mca_btl_openib_module_t* openib_btl,
struct ibv_qp* qp,
struct ibv_qp_attr* attr,
uint32_t lcl_psn,
uint32_t rem_qp_num,
uint32_t rem_psn,
uint16_t rem_lid,
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
uint32_t rem_mtu,
uint32_t port_num
)
{
attr->qp_state = IBV_QPS_RTR;
Bring over all the work from the /tmp/ib-hw-detect branch. In addition to my design and testing, it was conceptually approved by Gil, Gleb, Pasha, Brad, and Galen. Functionally [probably somewhat lightly] tested by Galen. We may still have to shake out some bugs during the next few months, but it seems to be working for all the cases that I can throw at it. Here's a summary of the changes from that branch: * Move MCA parameter registration to a new file (btl_openib_mca.c): * Properly check the retun status of registering MCA params * Check for valid values of MCA parameters * Make help strings better * Otherwise, the only default value of an MCA param that was changed was max_btls; it went from 4 to -1 (meaning: use all available) * Properly prototyped internal functions in _component.c * Made a bunch of functions static that didn't need to be public * Renamed to remove "mca_" prefix from static functions * Call new MCA param registration function * Call new INI file read/lookup/finalize functions * Updated a bunch of macros to be "BTL_" instead of "ORTE_" * Be a little more consistent with return values * Handle -1 for the max_btls MCA param * Fixed a free() that should have been an OBJ_RELEASE() * Some re-indenting * Added INI-file parsing * New flex file: btl_openib_ini.l * New default HCA params .ini file (probably to be expanded over time by other HCA vendors) * Added more show_help messages for parsing problems * Read in INI files and cache the values for later lookup * When component opens an HCA, lookup to see if any corresponding values were found in the INI files (ID'ed by the HCA vendor_id and vendor_part_id) * Added btl_openib_verbose MCA param that shows what the INI-file stuff does (e.g., shows which MTU your HCA ends up using) * Added btl_openib_hca_param_files as a colon-delimited list of INI files to check for values during startup (in order, left-to-right, just like the MCA base directory param). * MTU is currently the only value supported in this framework. * It is not a fatal error if we don't find params for the HCA in the INI file(s). Instead, just print a warning. New MCA param btl_openib_warn_no_hca_params_found can be used to disable printing the warning. * Add MTU to peer negotiation when making a connection * Exchange maximum MTU; select the lesser of the two This commit was SVN r11182.
2006-08-14 23:30:37 +04:00
attr->path_mtu = (openib_btl->hca->mtu < rem_mtu) ?
openib_btl->hca->mtu : rem_mtu;
if (mca_btl_openib_component.verbose) {
BTL_OUTPUT(("Set MTU to IBV value %d (%s bytes)", attr->path_mtu,
(attr->path_mtu == IBV_MTU_256) ? "256" :
(attr->path_mtu == IBV_MTU_512) ? "512" :
(attr->path_mtu == IBV_MTU_1024) ? "1024" :
(attr->path_mtu == IBV_MTU_2048) ? "2048" :
(attr->path_mtu == IBV_MTU_4096) ? "4096" :
"unknown (!)"));
}
attr->dest_qp_num = rem_qp_num;
attr->rq_psn = rem_psn;
attr->max_dest_rd_atomic = mca_btl_openib_component.ib_max_rdma_dst_ops;
attr->min_rnr_timer = mca_btl_openib_component.ib_min_rnr_timer;
attr->ah_attr.is_global = 0;
attr->ah_attr.dlid = rem_lid;
attr->ah_attr.sl = mca_btl_openib_component.ib_service_level;
attr->ah_attr.src_path_bits = openib_btl->src_path_bits;
attr->ah_attr.port_num = port_num;
/* JMS to be filled in later dynamically */
attr->ah_attr.static_rate = 0;
if(ibv_modify_qp(qp, attr,
IBV_QP_STATE |
IBV_QP_AV |
IBV_QP_PATH_MTU |
IBV_QP_DEST_QPN |
IBV_QP_RQ_PSN |
IBV_QP_MAX_DEST_RD_ATOMIC |
IBV_QP_MIN_RNR_TIMER)) {
BTL_ERROR(("error modifing QP to RTR errno says %s", strerror(errno)));
return OMPI_ERROR;
}
attr->qp_state = IBV_QPS_RTS;
attr->timeout = mca_btl_openib_component.ib_timeout;
attr->retry_cnt = mca_btl_openib_component.ib_retry_count;
attr->rnr_retry = mca_btl_openib_component.ib_rnr_retry;
attr->sq_psn = lcl_psn;
attr->max_rd_atomic = mca_btl_openib_component.ib_max_rdma_dst_ops;
if (ibv_modify_qp(qp, attr,
IBV_QP_STATE |
IBV_QP_TIMEOUT |
IBV_QP_RETRY_CNT |
IBV_QP_RNR_RETRY |
IBV_QP_SQ_PSN |
IBV_QP_MAX_QP_RD_ATOMIC)) {
BTL_ERROR(("error modifying QP to RTS errno says %s", strerror(errno)));
return OMPI_ERROR;
}
return OMPI_SUCCESS;
}
/**
* Return control fragment.
*/
static void mca_btl_openib_endpoint_credits(
mca_btl_base_module_t* btl,
struct mca_btl_base_endpoint_t* endpoint,
struct mca_btl_base_descriptor_t* descriptor,
int status)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int32_t checks, qp;
mca_btl_openib_frag_t *frag = (mca_btl_openib_frag_t*) descriptor;
qp = frag->qp_idx;
/* we don't acquire a wqe or token for credit message - so decrement */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OPAL_THREAD_ADD32(&endpoint->qps[qp].sd_wqe, -1);
/* check to see if there are additional credits to return */
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if((checks = OPAL_THREAD_ADD32(&endpoint->qps[qp].rd_pending_credit_chks,-1)) > 0) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].rd_pending_credit_chks, -checks);
if(btl_openib_check_send_credits(endpoint, qp)) {
mca_btl_openib_endpoint_send_credits(endpoint, qp);
}
}
}
/**
* Return credits to peer
*/
void mca_btl_openib_endpoint_send_credits(mca_btl_openib_endpoint_t* endpoint,
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
const int qp)
{
mca_btl_openib_module_t* openib_btl = endpoint->endpoint_btl;
mca_btl_openib_frag_t* frag;
mca_btl_openib_rdma_credits_header_t *credits_hdr;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
int do_rdma = 0, ib_rc;
int32_t cm_return;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
frag = endpoint->qps[qp].credit_frag;
if(NULL == frag) {
MCA_BTL_IB_FRAG_ALLOC_CREDIT_WAIT(openib_btl, frag, ib_rc);
frag->qp_idx = qp;
endpoint->qps[qp].credit_frag = frag;
}
assert(frag->qp_idx == qp);
credits_hdr =
(mca_btl_openib_rdma_credits_header_t*)frag->segment.seg_addr.pval;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if(BTL_OPENIB_EAGER_RDMA_QP(qp)) {
if(OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
} else {
do_rdma = 1;
}
}
if(0 == do_rdma) {
if(OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_sent, 1) >
(mca_btl_openib_component.qp_infos[qp].u.pp_qp.rd_rsv - 1)) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_sent, -1);
OPAL_THREAD_ADD32(&endpoint->qps[qp].rd_pending_credit_chks,
-endpoint->qps[qp].rd_pending_credit_chks);
return;
}
}
frag->base.des_cbfunc = mca_btl_openib_endpoint_credits;
frag->base.des_cbdata = NULL;
frag->endpoint = endpoint;
frag->hdr->tag = MCA_BTL_TAG_BTL;
GET_CREDITS(endpoint->qps[qp].u.pp_qp.rd_credits, frag->hdr->credits);
GET_CREDITS(endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
if(cm_return > 255) {
frag->hdr->cm_seen = 255;
cm_return -= 255;
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
} else {
frag->hdr->cm_seen = cm_return;
}
/* send eager RDMA credits only for eager RDMA QP */
if(BTL_OPENIB_EAGER_RDMA_QP(qp)) {
GET_CREDITS(endpoint->eager_rdma_local.credits,
credits_hdr->rdma_credits);
}
credits_hdr->control.type = MCA_BTL_OPENIB_CONTROL_CREDITS;
if(endpoint->nbo)
BTL_OPENIB_RDMA_CREDITS_HEADER_HTON((*credits_hdr));
frag->segment.seg_len = sizeof(mca_btl_openib_rdma_credits_header_t);
frag->sg_entry.addr = (unsigned long)frag->hdr;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
if((ib_rc = post_send(openib_btl, endpoint, frag, qp, do_rdma))) {
if(endpoint->nbo) {
BTL_OPENIB_HEADER_NTOH((*frag->hdr));
BTL_OPENIB_RDMA_CREDITS_HEADER_NTOH((*credits_hdr));
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OPAL_THREAD_ADD32(&endpoint->qps[qp].rd_pending_credit_chks, -1);
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.rd_credits, frag->hdr->credits);
OPAL_THREAD_ADD32(&endpoint->eager_rdma_local.credits, credits_hdr->rdma_credits);
if(do_rdma)
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
else
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_sent, -1);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
BTL_ERROR(("error posting send request errno %d says %s", ib_rc,
strerror(errno)));
}
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* local callback function for completion of eager rdma connect */
static void mca_btl_openib_endpoint_eager_rdma_connect_cb(
mca_btl_base_module_t* btl,
struct mca_btl_base_endpoint_t* endpoint,
struct mca_btl_base_descriptor_t* descriptor,
int status)
{
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
MCA_BTL_IB_FRAG_RETURN(((mca_btl_openib_module_t*)btl),
((mca_btl_openib_frag_t*)descriptor));
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* send the eager rdma conect message to the remote endpoint */
static int mca_btl_openib_endpoint_send_eager_rdma(
mca_btl_base_endpoint_t* endpoint)
{
mca_btl_openib_module_t* openib_btl = endpoint->endpoint_btl;
mca_btl_openib_eager_rdma_header_t *rdma_hdr;
mca_btl_openib_frag_t* frag;
int rc;
MCA_BTL_IB_FRAG_ALLOC_CREDIT_WAIT(openib_btl, frag, rc);
if(NULL == frag) {
return -1;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
frag->base.des_cbfunc = mca_btl_openib_endpoint_eager_rdma_connect_cb;
frag->base.des_cbdata = NULL;
frag->endpoint = endpoint;
frag->base.des_flags |= MCA_BTL_DES_FLAGS_PRIORITY;
frag->hdr->tag = MCA_BTL_TAG_BTL;
rdma_hdr = (mca_btl_openib_eager_rdma_header_t*)frag->segment.seg_addr.pval;
rdma_hdr->control.type = MCA_BTL_OPENIB_CONTROL_RDMA;
rdma_hdr->rkey = endpoint->eager_rdma_local.reg->mr->rkey;
rdma_hdr->rdma_start.lval = ompi_ptr_ptol(endpoint->eager_rdma_local.base.pval);
BTL_VERBOSE(("sending rkey %lu, rdma_start.lval %llu, pval %p, ival %u type %d and sizeof(rdma_hdr) %d\n",
rdma_hdr->rkey,
rdma_hdr->rdma_start.lval,
rdma_hdr->rdma_start.pval,
rdma_hdr->rdma_start.ival,
rdma_hdr->control.type,
sizeof(mca_btl_openib_eager_rdma_header_t)
));
frag->segment.seg_len = sizeof(mca_btl_openib_eager_rdma_header_t);
if(endpoint->nbo) {
BTL_OPENIB_EAGER_RDMA_CONTROL_HEADER_HTON((*rdma_hdr));
BTL_VERBOSE(("after HTON: sending rkey %lu, rdma_start.lval %llu, pval %p, ival %u\n",
rdma_hdr->rkey,
rdma_hdr->rdma_start.lval,
rdma_hdr->rdma_start.pval,
rdma_hdr->rdma_start.ival
));
}
if (mca_btl_openib_endpoint_send(endpoint, frag) != OMPI_SUCCESS) {
MCA_BTL_IB_FRAG_RETURN(openib_btl, frag);
BTL_ERROR(("Error sending RDMA buffer", strerror(errno)));
return -1;
}
return 0;
}
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
/* Setup eager RDMA buffers and notify the remote endpoint*/
void mca_btl_openib_endpoint_connect_eager_rdma(
mca_btl_openib_endpoint_t* endpoint)
{
mca_btl_openib_module_t* openib_btl = endpoint->endpoint_btl;
char *buf;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
mca_btl_openib_recv_frag_t *headers_buf;
unsigned int i;
orte_std_cntr_t index;
/* Set local rdma pointer to 1 temporarily so other threads will not try
* to enter the function */
if(!opal_atomic_cmpset_ptr(&endpoint->eager_rdma_local.base.pval, NULL,
(void*)1))
return;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
headers_buf = (mca_btl_openib_recv_frag_t*)
malloc(sizeof(mca_btl_openib_recv_frag_t) *
mca_btl_openib_component.eager_rdma_num);
if(NULL == headers_buf)
goto unlock_rdma_local;
buf = openib_btl->super.btl_mpool->mpool_alloc(openib_btl->super.btl_mpool,
openib_btl->eager_rdma_frag_size *
mca_btl_openib_component.eager_rdma_num,
mca_btl_openib_component.buffer_alignment,
MCA_MPOOL_FLAGS_CACHE_BYPASS,
(mca_mpool_base_registration_t**)&endpoint->eager_rdma_local.reg);
if(!buf)
goto free_headers_buf;
buf = buf + openib_btl->eager_rdma_frag_size -
sizeof(mca_btl_openib_footer_t) - openib_btl->super.btl_eager_limit -
sizeof(mca_btl_openib_header_t);
for(i = 0; i < mca_btl_openib_component.eager_rdma_num; i++) {
ompi_free_list_item_t *item;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
mca_btl_openib_recv_frag_t * frag;
mca_btl_openib_frag_init_data_t init_data;
item = (ompi_free_list_item_t*)&headers_buf[i];
item->registration = (void*)endpoint->eager_rdma_local.reg;
item->ptr = buf + i * openib_btl->eager_rdma_frag_size;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
OBJ_CONSTRUCT(item, mca_btl_openib_recv_frag_t);
init_data.length = mca_btl_openib_component.eager_limit;
init_data.order = mca_btl_openib_component.eager_rdma_qp;
init_data.type = MCA_BTL_OPENIB_FRAG_EAGER_RDMA;
init_data.list = NULL;
mca_btl_openib_frag_init(item, &init_data);
frag = (mca_btl_openib_recv_frag_t*) item;
frag->ftr = (mca_btl_openib_footer_t*)((char*)frag->segment.seg_addr.pval
+ frag->size);
MCA_BTL_OPENIB_RDMA_MAKE_REMOTE(frag->ftr);
((mca_btl_openib_frag_t*)item)->endpoint = endpoint;
}
endpoint->eager_rdma_local.frags = headers_buf;
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
endpoint->eager_rdma_local.rd_win =
mca_btl_openib_component.eager_rdma_num >> 2;
endpoint->eager_rdma_local.rd_win =
endpoint->eager_rdma_local.rd_win?endpoint->eager_rdma_local.rd_win:1;
/* set local rdma pointer to real value */
opal_atomic_cmpset_ptr(&endpoint->eager_rdma_local.base.pval, (void*)1,
buf);
if(mca_btl_openib_endpoint_send_eager_rdma(endpoint) == 0) {
/* This can never fail because max number of entries allocated
* at init time */
OBJ_RETAIN(endpoint);
This commit brings in two major things: 1. Galen's fine-grain control of queue pair resources in the openib BTL. 1. Pasha's new implementation of asychronous HCA event handling. Pasha's new implementation doesn't take much explanation, but the new "multifrag" stuff does. Note that "svn merge" was not used to bring this new code from the /tmp/ib_multifrag branch -- something Bad happened in the periodic trunk pulls on that branch making an actual merge back to the trunk effectively impossible (i.e., lots and lots of arbitrary conflicts and artifical changes). :-( == Fine-grain control of queue pair resources == Galen's fine-grain control of queue pair resources to the OpenIB BTL (thanks to Gleb for fixing broken code and providing additional functionality, Pasha for finding broken code, and Jeff for doing all the svn work and regression testing). Prior to this commit, the OpenIB BTL created two queue pairs: one for eager size fragments and one for max send size fragments. When the use of the shared receive queue (SRQ) was specified (via "-mca btl_openib_use_srq 1"), these QPs would use a shared receive queue for receive buffers instead of the default per-peer (PP) receive queues and buffers. One consequence of this design is that receive buffer utilization (the size of the data received as a percentage of the receive buffer used for the data) was quite poor for a number of applications. The new design allows multiple QPs to be specified at runtime. Each QP can be setup to use PP or SRQ receive buffers as well as giving fine-grained control over receive buffer size, number of receive buffers to post, when to replenish the receive queue (low water mark) and for SRQ QPs, the number of outstanding sends can also be specified. The following is an example of the syntax to describe QPs to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues: {{{ -mca btl_openib_receive_queues \ "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32" }}} Each QP description is delimited by ";" (semicolon) with individual fields of the QP description delimited by "," (comma). The above example therefore describes 4 QPs. The first QP is: P,128,16,4 Meaning: per-peer receive buffer QPs are indicated by a starting field of "P"; the first QP (shown above) is therefore a per-peer based QP. The second field indicates the size of the receive buffer in bytes (128 bytes). The third field indicates the number of receive buffers to allocate to the QP (16). The fourth field indicates the low watermark for receive buffers at which time the BTL will repost receive buffers to the QP (4). The second QP is: S,1024,256,128,32 Shared receive queue based QPs are indicated by a starting field of "S"; the second QP (shown above) is therefore a shared receive queue based QP. The second, third and fourth fields are the same as in the per-peer based QP. The fifth field is the number of outstanding sends that are allowed at a given time on the QP (32). This provides a "good enough" mechanism of flow control for some regular communication patterns. QPs MUST be specified in ascending receive buffer size order. This requirement may be removed prior to 1.3 release. This commit was SVN r15474.
2007-07-18 05:15:59 +04:00
assert(((opal_object_t*)endpoint)->obj_reference_count == 2);
orte_pointer_array_add(&index, openib_btl->eager_rdma_buffers,
endpoint);
/* from this point progress function starts to poll new buffer */
OPAL_THREAD_ADD32(&openib_btl->eager_rdma_buffers_count, 1);
return;
}
openib_btl->super.btl_mpool->mpool_free(openib_btl->super.btl_mpool,
buf, (mca_mpool_base_registration_t*)endpoint->eager_rdma_local.reg);
free_headers_buf:
free(headers_buf);
unlock_rdma_local:
/* set local rdma pointer back to zero. Will retry later */
opal_atomic_cmpset_ptr(&endpoint->eager_rdma_local.base.pval,
endpoint->eager_rdma_local.base.pval, NULL);
endpoint->eager_rdma_local.frags = NULL;
}