Skip to content

pmodels/armci-mpi

Repository files navigation

Authors

  • James Dinan (MPI-2 implementation)
  • Jeff Hammond (MPI-3 implementation)

Introduction

This project provides a full, high performance, portable implementation of the ARMCI runtime system using MPI's remote memory access (RMA) functionality.

Quality Assurance

Build Status

See Travis for failure details. All recent failures have been caused by dependencies (system toolchain or MPI library).

Installing Only ARMCI-MPI

ARMCI-MPI uses autoconf and must be configured before compiling:

 $ ./configure

Many configure options are provided, run configure --help for details. After configuring the source tree, the code can be built and installed by running:

 $ make && make install

MPI Library Issues

The quality of MPI-RMA implementations varies. We recommend that you always use the absolute latest release or release candidate version of MPI unless you are aware of a specific issues that prevents this.

With the exception of the IBM Blue Gene platforms, all MPI libraries known to the ARMCI-MPI developers support MPI-3 RMA, hence the MPI-2 RMA implementation has been deprecated and is no longer supported. If you are using ARMCI-MPI on a Blue Gene system, use the legacy branch or contact the developers for assistance.

MPI-3

As of September 2018, MPICH 3.3b3 and Open-MPI 3.1.2 are passing all of the tests in Travis CI. Information about other implementations will be added here soon.

As of April 2014, the following implementations were known to work correctly with ARMCI-MPI (MPI-3 version):

  • MPICH 3.0.4 and later on Mac, Linux SMPs and SGI SMPs.
  • MVAPICH2 2.0a and later on Linux InfiniBand clusters.
  • CrayMPI 6.1.0 and later on Cray XC30.
  • SGI MPT 2.09 on SGI SMPs.
  • Open-MPI development version on Mac (set ARMCI_STRIDED_METHOD=IOV and ARMCI_IOV_METHOD=BATCHED)

Note that a bug in MPICH 3.0 or 3.1 that propagated to MVAPICH2, Cray MPI and Intel MPI affects correctness when windows are backed by shared-memory. This bug affects ARMCI_Rmw and is avoided by setting ARMCI_USE_WIN_ALLOCATE=0 in your runtime environment. This may negatively affect performance in some cases and prevents one from using Casper, hence is not the default.

MPI-2

As of August, 2011 the following MPI-2 implementations were known to work correctly with ARMCI-MPI (MPI-2 version):

  • MPICH2 and MPICH 3+
  • MVAPICH2 1.6
  • Cray MPI on Cray XE6
  • IBM MPI on BG/P (set ARMCI_STRIDED_METHOD=IOV and ARMCI_IOV_METHOD=BATCHED for performance reasons)
  • Open-MPI 1.5.4 (set ARMCI_STRIDED_METHOD=IOV and ARMCI_IOV_METHOD=BATCHED for correctness reasons)

The following MPI-2 implementations are known to fail with ARMCI-MPI:

  • MVAPICH2 prior to 1.6

Installing Global Arrays with ARMCI-MPI

To build GA (version 5.2 or later) with ARMCI-MPI (any version), use the configure option --with-armci=$(PATH_TO_ARMCI_MPI) and make sure that you use the same MPI implementation with GA that was used to compile ARMCI-MPI.

ARMCI-MPI (MPI-3) has been tested extensively with GA since version 5.2.

Installing NWChem with ARMCI-MPI

If you are an NWChem user, you can use ${NWCHEM_TOP}/src/tools/install-armci-mpi without having to download or build ARMCI-MPI manually.

The ARMCI-MPI Test Suite

ARMCI-MPI includes a set of testing and benchmark programs located under tests/ and benchmarks/. These programs can be compiled and run via:

$ make check MPIEXEC="mpiexec -n 4"

The MPIEXEC variable is optional and is used to override the default MPI launch command. If you want only to build the test suite, the following target can be used:

$ make checkprogs

ARMCI-MPI Errata

Direct access to local buffers

  • Because of MPI-2's semantics, you are not allowed to access shared memory directly, it must be through put/get. Alternatively you can use the new ARMCI_Access_begin/end() functions.

  • MPI-3 allows direct access provided one uses a synchronization operation afterwards. The ARMCI_Access_begin/end() functions are also valid.

Progress semantics

  • On some MPI implementations and networks you may need to enable implicit progress. In many cases this is done through an environment variable. For MPICH2: set MPICH_ASYNC_PROGRESS; for MVAPICH2 recompile with --enable-async-progress and set MPICH_ASYNC_PROGRESS; set DCMF_INTERRUPTS=1 for MPI on BGP; etc.

See this page for more information on activating asynchronous progress in MPI. However, we find that most platforms show no improvement and often a decrease in performance, provided the application makes calls to GA/ARMCI/MPI frequently enough on all processes.

We recommend the user of Casper for asynchronous progress in ARMCI-MPI. See the Casper website for details.

Environment Variables:

Boolean environment variables are enabled when set to a value beginning with 't', 'T', 'y', 'Y', or '1'; any other value is interpreted as false.

Debugging Options

ARMCI_VERBOSE (boolean)

Enable extra status output from ARMCI-MPI.

ARMCI_DEBUG_ALLOC (boolean)

Turn on extra shared allocation debugging.

ARMCI_FLUSH_BARRIERS (boolean) (deprecated)

Enable/disable extra communication flushing in ARMCI_Barrier. Extra flushes are present to help make unsafe DLA safer. (This option is deprecated with the ARMCI-MPI3 implementation.)

Performance Options

ARMCI_CACHE_RANK_TRANSLATION (boolean)

Create a table to more quickly translate between absolute and group ranks.

ARMCI_PROGRESS_THREAD (boolean)

Create a Pthread to poke the MPI progress engine.

ARMCI_PROGRESS_USLEEP (int)

Argument to usleep() to pause the progress polling loop.

Noncollective Groups

ARMCI_NONCOLLECTIVE_GROUPS (boolean)

Enable noncollective ARMCI group formation; group creation is collective on the output group rather than the parent group.

Shared Buffer Protection

ARMCI_SHR_BUF_METHOD = { COPY (default), NOGUARD }

ARMCI policy for managing shared origin buffers in communication operations: lock the buffer (unsafe, but fast), copy the buffer (safe), or don't guard the buffer - assume that the system is cache coherent and MPI supports unlocked load/store.

Strided Options

ARMCI_STRIDED_METHOD = { DIRECT (default), IOV }

Select the method for processing strided operations.

I/O Vector Options

ARMCI_IOV_METHOD = { AUTO (default), CONSRV, BATCHED, DIRECT }

Select the IO vector communication strategy: automatic; a "conservative" implementation that does lock/unlock around each operation; an implementation that issues batches of operations within a single lock/unlock epoch; and a direct implementation that generates datatypes for the origin and target and issues a single operation using them.

ARMCI_IOV_CHECKS (boolean)

Enable (expensive) IOV safety/debugging checks (not recommended for performance runs).

ARMCI_IOV_BATCHED_LIMIT = { 0 (default), 1, ... }

Set the maximum number of one-sided operations per epoch for the BATCHED IOV method. Zero (default) is unlimited.