611d7f9f6b
This creates a really bad scaling behavior. Users have found a nearly 20% launch time differential between mpirun and PMI, with PMI being the slower method. Some of the problem is attributable to poor exchange algorithms in RM's like Slurm and Alps, but we make things worse by calling "get" so many times. Nathan (with a tad advice from me) has attempted to alleviate this problem by reducing the number of "get" calls. This required the following changes: * upon first request for data, have the OPAL db pmi component fetch and decode *all* the info from a given remote proc. It turned out we weren't caching the info, so we would continually request it and only decode the piece we needed for the immediate request. We now decode all the info and push it into the db hash component for local storage - and then all subsequent retrievals are fulfilled locally * reduced the amount of data by eliminating the exchange of the OMPI_ARCH value if heterogeneity is not enabled. This was used solely as a check so we would error out if the system wasn't actually homogeneous, which was fine when we thought there was no cost in doing the check. Unfortunately, at large scale and with direct launch, there is a non-zero cost of making this test. We are open to finding a compromise (perhaps turning the test off if requested?), if people feel strongly about performing the test * reduced the amount of RTE data being automatically fetched, and fetched the rest only upon request. In particular, we no longer immediately fetch the hostname (which is only used for error reporting), but instead get it when needed. Likewise for the RML uri as that info is only required for some (not all) environments. In addition, we no longer fetch the locality unless required, relying instead on the PMI clique info to tell us who is on our local node (if additional info is required, the fetch is performed when a modex_recv is issued). Again, all this only impacts direct launch - all the info is provided when launched via mpirun as there is no added cost to getting it Barring objections, we may move this (plus any required other pieces) to the 1.7 branch once it soaks for an appropriate time. This commit was SVN r29040.
196 строки
6.9 KiB
C
196 строки
6.9 KiB
C
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
|
/*
|
|
* Copyright (c) 2011-2012 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* Copyright (c) 2011-2013 UT-Battelle, LLC. All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "btl_ugni.h"
|
|
|
|
#include "btl_ugni_endpoint.h"
|
|
#include "btl_ugni_smsg.h"
|
|
|
|
static void mca_btl_ugni_ep_construct (mca_btl_base_endpoint_t *ep)
|
|
{
|
|
memset ((char *) ep + sizeof(ep->super), 0, sizeof (*ep) - sizeof (ep->super));
|
|
OBJ_CONSTRUCT(&ep->frag_wait_list, opal_list_t);
|
|
OBJ_CONSTRUCT(&ep->lock, opal_mutex_t);
|
|
}
|
|
|
|
static void mca_btl_ugni_ep_destruct (mca_btl_base_endpoint_t *ep)
|
|
{
|
|
OBJ_DESTRUCT(&ep->frag_wait_list);
|
|
OBJ_DESTRUCT(&ep->lock);
|
|
}
|
|
|
|
OBJ_CLASS_INSTANCE(mca_btl_base_endpoint_t, opal_list_item_t,
|
|
mca_btl_ugni_ep_construct, mca_btl_ugni_ep_destruct);
|
|
|
|
static inline int mca_btl_ugni_ep_smsg_get_mbox (mca_btl_base_endpoint_t *ep) {
|
|
mca_btl_ugni_module_t *ugni_module = ep->btl;
|
|
ompi_free_list_item_t *mbox;
|
|
|
|
OMPI_FREE_LIST_GET_MT(&ugni_module->smsg_mboxes, mbox);
|
|
if (OPAL_UNLIKELY(NULL == mbox)) {
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
}
|
|
|
|
ep->mailbox = (mca_btl_ugni_smsg_mbox_t *) mbox;
|
|
|
|
/* per ugni spec we need to zero mailbox data before connecting */
|
|
memset ((char *)ep->mailbox->smsg_attrib.msg_buffer + ep->mailbox->smsg_attrib.mbox_offset, 0,
|
|
ep->mailbox->smsg_attrib.buff_size);
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
int mca_btl_ugni_ep_disconnect (mca_btl_base_endpoint_t *ep, bool send_disconnect) {
|
|
gni_return_t rc;
|
|
|
|
do {
|
|
if (MCA_BTL_UGNI_EP_STATE_INIT == ep->state) {
|
|
/* nothing to do */
|
|
break;
|
|
}
|
|
|
|
if (MCA_BTL_UGNI_EP_STATE_CONNECTED == ep->state && send_disconnect) {
|
|
rc = GNI_SmsgSendWTag (ep->smsg_ep_handle, NULL, 0, NULL, 0, -1,
|
|
MCA_BTL_UGNI_TAG_DISCONNECT);
|
|
if (GNI_RC_SUCCESS != rc) {
|
|
BTL_VERBOSE(("btl/ugni could not send close message"));
|
|
}
|
|
|
|
/* we might want to wait for local completion here (do we even care) */
|
|
}
|
|
|
|
(void) ompi_common_ugni_ep_destroy (&ep->smsg_ep_handle);
|
|
(void) ompi_common_ugni_ep_destroy (&ep->rdma_ep_handle);
|
|
|
|
OMPI_FREE_LIST_RETURN_MT(&ep->btl->smsg_mboxes, ((ompi_free_list_item_t *) ep->mailbox));
|
|
ep->mailbox = NULL;
|
|
|
|
ep->state = MCA_BTL_UGNI_EP_STATE_INIT;
|
|
} while (0);
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static inline int mca_btl_ugni_ep_connect_start (mca_btl_base_endpoint_t *ep) {
|
|
int rc;
|
|
|
|
/* get the modex info for this endpoint and setup a ugni endpoint */
|
|
rc = ompi_common_ugni_endpoint_for_proc (ep->btl->device, ep->peer_proc, &ep->common);
|
|
if (OMPI_SUCCESS != rc) {
|
|
assert (0);
|
|
return rc;
|
|
}
|
|
|
|
BTL_VERBOSE(("initiaiting connection to remote peer with address: %u id: %u",
|
|
ep->common->ep_rem_addr, ep->common->ep_rem_id));
|
|
|
|
/* bind endpoint to remote address */
|
|
/* we bind two endpoints to seperate out local smsg completion and local fma completion */
|
|
rc = ompi_common_ugni_ep_create (ep->common, ep->btl->smsg_local_cq, &ep->smsg_ep_handle);
|
|
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
|
|
return rc;
|
|
}
|
|
|
|
rc = ompi_common_ugni_ep_create (ep->common, ep->btl->rdma_local_cq, &ep->rdma_ep_handle);
|
|
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
|
|
return rc;
|
|
}
|
|
|
|
/* build connection data */
|
|
rc = mca_btl_ugni_ep_smsg_get_mbox (ep);
|
|
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
|
|
return rc;
|
|
}
|
|
|
|
ep->state = MCA_BTL_UGNI_EP_STATE_CONNECTING;
|
|
|
|
memset (&ep->remote_smsg_attrib, 0, sizeof (ep->remote_smsg_attrib));
|
|
|
|
BTL_VERBOSE(("btl/ugni connection to remote peer initiated"));
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static inline int mca_btl_ugni_ep_connect_finish (mca_btl_base_endpoint_t *ep) {
|
|
gni_return_t grc;
|
|
int rc;
|
|
|
|
BTL_VERBOSE(("finishing connection. remote attributes: msg_type = %d, msg_buffer = %p, buff_size = %d, "
|
|
"mem_hndl = {qword1 = %" PRIu64 ", qword2 = %" PRIu64 "}, mbox = %d, mbox_maxcredit = %d, "
|
|
"msg_maxsize = %d", ep->remote_smsg_attrib.msg_type, ep->remote_smsg_attrib.msg_buffer,
|
|
ep->remote_smsg_attrib.buff_size, ep->remote_smsg_attrib.mem_hndl.qword1,
|
|
ep->remote_smsg_attrib.mem_hndl.qword2, ep->remote_smsg_attrib.mbox_offset,
|
|
ep->remote_smsg_attrib.mbox_maxcredit, ep->remote_smsg_attrib.msg_maxsize));
|
|
|
|
BTL_VERBOSE(("finishing connection. local attributes: msg_type = %d, msg_buffer = %p, buff_size = %d, "
|
|
"mem_hndl = {qword1 = %" PRIu64 ", qword2 = %" PRIu64 "}, mbox = %d, mbox_maxcredit = %d, "
|
|
"msg_maxsize = %d", ep->mailbox->smsg_attrib.msg_type, ep->mailbox->smsg_attrib.msg_buffer,
|
|
ep->mailbox->smsg_attrib.buff_size, ep->mailbox->smsg_attrib.mem_hndl.qword1,
|
|
ep->mailbox->smsg_attrib.mem_hndl.qword2, ep->mailbox->smsg_attrib.mbox_offset,
|
|
ep->mailbox->smsg_attrib.mbox_maxcredit, ep->mailbox->smsg_attrib.msg_maxsize));
|
|
|
|
grc = GNI_SmsgInit (ep->smsg_ep_handle, &ep->mailbox->smsg_attrib, &ep->remote_smsg_attrib);
|
|
if (OPAL_UNLIKELY(GNI_RC_SUCCESS != grc)) {
|
|
BTL_ERROR(("error initializing SMSG protocol. rc = %d", grc));
|
|
|
|
return ompi_common_rc_ugni_to_ompi (grc);
|
|
}
|
|
|
|
ep->state = MCA_BTL_UGNI_EP_STATE_CONNECTED;
|
|
|
|
/* send all pending messages */
|
|
BTL_VERBOSE(("endpoint connected. posting %u sends", (unsigned int) opal_list_get_size (&ep->frag_wait_list)));
|
|
|
|
rc = mca_btl_progress_send_wait_list (ep);
|
|
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
|
|
ep->wait_listed = true;
|
|
opal_list_append (&ep->btl->ep_wait_list, &ep->super);
|
|
}
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static inline int mca_btl_ugni_directed_ep_post (mca_btl_base_endpoint_t *ep) {
|
|
gni_return_t rc;
|
|
|
|
rc = GNI_EpPostDataWId (ep->smsg_ep_handle, &ep->mailbox->smsg_attrib, sizeof (ep->mailbox->smsg_attrib),
|
|
&ep->remote_smsg_attrib, sizeof (ep->remote_smsg_attrib),
|
|
MCA_BTL_UGNI_CONNECT_DIRECTED_ID | ep->common->ep_rem_id);
|
|
|
|
return ompi_common_rc_ugni_to_ompi (rc);
|
|
}
|
|
|
|
int mca_btl_ugni_ep_connect_progress (mca_btl_base_endpoint_t *ep) {
|
|
int rc;
|
|
|
|
if (MCA_BTL_UGNI_EP_STATE_CONNECTED == ep->state) {
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
if (MCA_BTL_UGNI_EP_STATE_INIT == ep->state) {
|
|
rc = mca_btl_ugni_ep_connect_start (ep);
|
|
if (OMPI_SUCCESS != rc) {
|
|
return rc;
|
|
}
|
|
}
|
|
|
|
if (GNI_SMSG_TYPE_INVALID == ep->remote_smsg_attrib.msg_type) {
|
|
/* use datagram to exchange connection information with the remote peer */
|
|
rc = mca_btl_ugni_directed_ep_post (ep);
|
|
if (OMPI_SUCCESS == rc) {
|
|
rc = OMPI_ERR_RESOURCE_BUSY;
|
|
}
|
|
return rc;
|
|
}
|
|
|
|
return mca_btl_ugni_ep_connect_finish (ep);
|
|
}
|