OFI MTL:
--------
The OFI MTL supports Libfabric (a.k.a. Open Fabrics Interfaces OFI,
https://ofiwg.github.io/libfabric/) tagged APIs (fi_tagged(3)). At
initialization time, the MTL queries libfabric for providers supporting tag matching
(fi_getinfo(3)). Libfabric will return a list of providers that satisfy the requested
capabilities, having the most performant one at the top of the list.
The user may modify the OFI provider selection with mca parameters
mtl_ofi_provider_include or mtl_ofi_provider_exclude.
PROGRESS:
---------
The MTL registers a progress function to opal_progress. There is currently
no support for asynchronous progress. The progress function reads multiple events
from the OFI provider Completion Queue (CQ) per iteration (defaults to 100, can be
modified with the mca mtl_ofi_progress_event_cnt) and iterates until the
completion queue is drained.
COMPLETIONS:
------------
Each operation uses a request type ompi_mtl_ofi_request_t which includes a reference
to an operation specific completion callback, an MPI request, and a context. The
context (fi_context) is used to map completion events with MPI_requests when reading the
CQ.
OFI TAG:
--------
MPI needs to send 96 bits of information per message (32 bits communicator id,
32 bits source rank, 32 bits MPI tag) but OFI only offers 64 bits tags. In
addition, the OFI MTL uses 2 bits of the OFI tag for the synchronous send protocol.
Therefore, there are only 62 bits available in the OFI tag for message usage. The
OFI MTL offers the mtl_ofi_tag_mode mca parameter with 4 modes to address this:
"auto" (Default):
After the OFI provider is selected, a runtime check is performed to assess
FI_REMOTE_CQ_DATA and FI_DIRECTED_RECV support (see fi_tagged(3), fi_msg(2)
and fi_getinfo(3)). If supported, "ofi_tag_full" is used. If not supported,
fall back to "ofi_tag_1".
"ofi_tag_1":
For providers that do not support FI_REMOTE_CQ_DATA, the OFI MTL will
trim the fields (Communicator ID, Source Rank, MPI tag) to make them fit the 62
bits available bit in the OFI tag. There are two options available with different
number of bits for the Communicator ID and MPI tag fields. This tag distribution
offers: 12 bits for Communicator ID (max Communicator ID 4,095) subject to
provider reserved bits (see mem_tag_format below), 18 bits for Source Rank (max
Source Rank 262,143), 32 bits for MPI tag (max MPI tag is INT_MAX).
"ofi_tag_2":
Same as 2 "ofi_tag_1" but offering a different OFI tag distribution for
applications that may require a greater number of supported Communicators at the
expense of fewer MPI tag bits. This tag distribution offers: 24 bits for
Communicator ID (max Communicator ED 16,777,215. See mem_tag_format below), 18
bits for Source Rank (max Source Rank 262,143), 20 bits for MPI tag (max MPI tag
524,287).
"ofi_tag_full":
For executions that cannot accept trimming source rank or MPI tag, this mode sends
source rank for each message in the CQ DATA. The Source Rank is made available at
the remote process CQ (FI_CQ_FORMAT_TAGGED is used, see fi_cq(3)) at the completion
of the matching receive operation. Since the minimum size for FI_REMOTE_CQ_DATA
is 32 bits, the Source Rank fits with no limitations. The OFI tag is used for the
Communicator id (28 bits, max Communicator ID 268,435,455. See mem_tag_format below),
and the MPI tag (max MPI tag is INT_MAX). If this mode is selected by the user
and FI_REMOTE_CQ_DATA or FI_DIRECTED_RECV are not supported, the execution will abort.
mem_tag_format (fi_endpoint(3))
Some providers can reserve the higher order bits from the OFI tag for internal purposes.
This is signaled in mem_tag_format (see fi_endpoint(3)) by setting higher order bits
to zero. In such cases, the OFI MTL will reduce the number of communicator ids supported
by reducing the bits available for the communicator ID field in the OFI tag.
SCALABLE ENDPOINTS:
-------------------
OFI MTL supports OFI Scalable Endpoints (SEP) feature as a means to improve
multi-threaded application throughput and message rate. Currently the feature
is designed to utilize multiple TX/RX contexts exposed by the OFI provider in
conjunction with a multi-communicator MPI application model. Therefore, new OFI
contexts are created as and when communicators are duplicated in a lazy fashion
instead of creating them all at once during init time and this approach also
favours only creating as many contexts as needed.
1. Multi-communicator model:
With this approach, the MPI application is requried to first duplicate
the communicators it wants to use with MPI operations (ideally creating
as many communicators as the number of threads it wants to use to call
into MPI). The duplicated communicators are then used by the
corresponding threads to perform MPI operations. A possible usage
scenario could be in an MPI + OMP application as follows
(example limited to 2 ranks):
MPI_Comm dup_comm[n];
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
for (i = 0; i < n; i++) {
MPI_Comm_dup(MPI_COMM_WORLD, &dup_comm[i]);
}
if (rank == 0) {
#pragma omp parallel for private(host_sbuf, host_rbuf) num_threads(n)
for (i = 0; i < n ; i++) {
MPI_Send(host_sbuf, MYBUFSIZE, MPI_CHAR,
1, MSG_TAG, dup_comm[i]);
MPI_Recv(host_rbuf, MYBUFSIZE, MPI_CHAR,
1, MSG_TAG, dup_comm[i], &status);
}
} else if (rank == 1) {
#pragma omp parallel for private(status, host_sbuf, host_rbuf) num_threads(n)
for (i = 0; i < n ; i++) {
MPI_Recv(host_rbuf, MYBUFSIZE, MPI_CHAR,
0, MSG_TAG, dup_comm[i], &status);
MPI_Send(host_sbuf, MYBUFSIZE, MPI_CHAR,
0, MSG_TAG, dup_comm[i]);
}
}
2. MCA variables:
To utilize the feature, the following MCA variables need to be set:
mtl_ofi_enable_sep:
This MCA variable needs to be set to enable the use of Scalable Endpoints (SEP)
feature in the OFI MTL. The underlying provider is also checked to ensure the
feature is supported. If the provider chosen does not support it, user needs
to either set this variable to 0 or select a different provider which supports
the feature.
For single-threaded applications one OFI context is sufficient, so OFI SEPs
may not add benefit.
Note that mtl_ofi_thread_grouping (see below) needs to be enabled to use the
different OFI SEP contexts. Otherwise, only one context (ctxt 0) will be used.
Default: 0
Command-line syntax:
"-mca mtl_ofi_enable_sep 1"
mtl_ofi_thread_grouping:
Turn Thread Grouping feature on. This is needed to use the Multi-communicator
model explained above. This means that the OFI MTL will use the communicator
ID to decide the SEP contexts to be used by the thread. In this way, each
thread will have direct access to different OFI resources. If disabled,
only context 0 will be used.
Requires mtl_ofi_enable_sep to be set to 1.
Default: 0
It is not recommended to set the MCA variable for:
- Multi-threaded MPI applications not following multi-communicator approach.
- Applications that have multiple threads using a single communicator as
it may degrade performance.
Command-line syntax:
"-mca mtl_ofi_thread_grouping 1"
mtl_ofi_num_ctxts:
This MCA variable allows user to set the number of OFI SEP contexts the
application expects to use. For multi-threaded applications using Thread
Grouping feature, this number should be set to the number of user threads
that will call into MPI. This variable will only have effect if
mtl_ofi_enable_sep is set to 1.
Default: 1
Command-line syntax:
"-mca mtl_ofi_num_ctxts N" [ N: number of OFI contexts required by
application ]
3. Notes on performance:
- OFI MTL will create as many TX/RX contexts as set by MCA mtl_ofi_num_ctxts.
The number of contexts that can be created is also limited by the underlying
provider as each provider may have different thresholds. Once the threshold
is exceeded, contexts are used in a round-robin fashion which leads to
resource sharing among threads. Therefore locks are required to guard
against race conditions. For performance, it is recommended to have
Number of threads = Number of communicators = Number of contexts
For example, when using PSM2 provider, the number of contexts is dictated
by the Intel Omni-Path HFI1 driver module.
- OPAL layer allows for multiple threads to enter progress simultaneously. To
enable this feature, user needs to set MCA variable
"max_thread_in_progress". When using Thread Grouping feature, it is
recommended to set this MCA parameter to the number of threads expected to
call into MPI as it provides performance benefits.
Command-line syntax:
"-mca opal_max_thread_in_progress N" [ N: number of threads expected to
make MPI calls ]
Default: 1
- For applications using a single thread with multiple communicators and MCA
variable "mtl_ofi_thread_grouping" set to 1, the MTL will use multiple
contexts, but the benefits may be negligible as only one thread is driving
progress.
SPECIALIZED FUNCTIONS:
-------------------
To improve performance when calling message passing APIs in the OFI mtl
specialized functions are generated at compile time that eliminate all the
if conditionals that can be determined at init and don't need to be
queried again during the critical path. These functions are generated by
perl scripts during make which generate functions and symbols for every
combination of flags for each function.
1. ADDING NEW FLAGS FOR SPECIALIZATION OF EXISTING FUNCTION:
To add a new flag to an existing specialized function for handling cases
where different OFI providers may or may not support a particular feature,
then you must follow these steps:
1) Update the "_generic" function in mtl_ofi.h with the new flag and
the if conditionals to read the new value.
2) Update the *.pm file corresponding to the function with the new flag in:
gen_funcs(), gen_*_function(), & gen_*_sym_init()
3) Update mtl_ofi_opt.h with:
The new flag as #define NEW_FLAG_TYPES #NUMBER_OF_STATES
example: #define OFI_CQ_DATA 2 (only has TRUE/FALSE states)
Update the function's types with:
#define OMPI_MTL_OFI_FUNCTION_TYPES [NEW_FLAG_TYPES]
2. ADDING A NEW FUNCTION FOR SPECIALIZATION:
To add a new function to be specialized you must
follow these steps:
1) Create a new mtl_ofi_"function_name"_opt.pm based off opt_common/mtl_ofi_opt.pm.template
2) Add new .pm file to generated_source_modules in Makefile.am
3) Add .c file to generated_sources in Makefile.am named the same as the corresponding .pm file
4) Update existing or create function in mtl_ofi.h to _generic with new flags.
5) Update mtl_ofi_opt.h with:
a) New function types: #define OMPI_MTL_OFI_FUNCTION_TYPES [FLAG_TYPES]
b) Add new function to the struct ompi_mtl_ofi_symtable:
struct ompi_mtl_ofi_symtable {
...
int (*ompi_mtl_ofi_FUNCTION OMPI_MTL_OFI_FUNCTION_TYPES )
}
c) Add new symbol table init function definition:
void ompi_mtl_ofi_FUNCTION_symtable_init(struct ompi_mtl_ofi_symtable* sym_table);
6) Add calls to init the new function in the symbol table and assign the function
pointer to be used based off the flags in mtl_ofi_component.c:
ompi_mtl_ofi_FUNCTION_symtable_init(&ompi_mtl_ofi.sym_table);
ompi_mtl_ofi.base.mtl_FUNCTION =
ompi_mtl_ofi.sym_table.ompi_mtl_ofi_FUNCTION[ompi_mtl_ofi.flag];
3. EXAMPLE SPECIALIZED FILE:
The code below is an example of what is generated by the specialization
scripts for use in the OFI mtl. This code specializes the blocking
send functionality based on FI_REMOTE_CQ_DATA & OFI Scalable Endpoint support
provided by an OFI Provider. Only one function and symbol is used during
runtime based on if FI_REMOTE_CQ_DATA is supported and/or if OFI Scalable
Endpoint support is enabled.
/*
* Copyright (c) 2013-2018 Intel, Inc. All rights reserved
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "mtl_ofi.h"
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_false_false(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
const bool OFI_CQ_DATA = false;
const bool OFI_SCEP_EPS = false;
return ompi_mtl_ofi_send_generic(mtl, comm, dest, tag,
convertor, mode,
OFI_CQ_DATA, OFI_SCEP_EPS);
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_false_true(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
const bool OFI_CQ_DATA = false;
const bool OFI_SCEP_EPS = true;
return ompi_mtl_ofi_send_generic(mtl, comm, dest, tag,
convertor, mode,
OFI_CQ_DATA, OFI_SCEP_EPS);
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_true_false(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
const bool OFI_CQ_DATA = true;
const bool OFI_SCEP_EPS = false;
return ompi_mtl_ofi_send_generic(mtl, comm, dest, tag,
convertor, mode,
OFI_CQ_DATA, OFI_SCEP_EPS);
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_true_true(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
const bool OFI_CQ_DATA = true;
const bool OFI_SCEP_EPS = true;
return ompi_mtl_ofi_send_generic(mtl, comm, dest, tag,
convertor, mode,
OFI_CQ_DATA, OFI_SCEP_EPS);
}
void ompi_mtl_ofi_send_symtable_init(struct ompi_mtl_ofi_symtable* sym_table)
{
sym_table->ompi_mtl_ofi_send[false][false]
= ompi_mtl_ofi_send_false_false;
sym_table->ompi_mtl_ofi_send[false][true]
= ompi_mtl_ofi_send_false_true;
sym_table->ompi_mtl_ofi_send[true][false]
= ompi_mtl_ofi_send_true_false;
sym_table->ompi_mtl_ofi_send[true][true]
= ompi_mtl_ofi_send_true_true;
}
###
