mercury-hpc / fabtsuite

libfabric test suite

fabtsuite: a libfabric test suite

This is a test suite for libfabric designed to exercise features having particular interest for the Mochi project.

Project contents

The major constituents of this project are the test program under transfer/, a test script under scripts/, and the project documentation under doc/.

The test program, which lives under transfer/, assumes either a server or client personality, depending on the name by which it is invoked. Invoked as fabtget, it is the test server, and as fabtput, the test client.

scripts/fabtrun is the main test script.

doc/building.md, doc/building_cmake.md, and doc/building_spack.md tell how to build the project.

doc/tests.md describes the tests in this suite. doc/tests_mn.md describes how to run test using multiple nodes.

Recipes for building and executing test on specific platforms

• OLCF Crusher

fabtsuite's Issues

add documentation links in README.md

The top-level README.md names some other markdown files that contain documentation.

Can we turn these into links so that they are easier to browse/follow online from the github repository?

The syntax is [LINK_TEXT](RELATIVE_PATH_TO_ANOTHER_MD_FILE)

documentation for ctest runs

If I understand correctly, there is a mode of operation in which you execute a single script (on a login node?) and it submits multiple scheduler jobs and reports the results from each. Is this mode documented yet?

Browsing the various .md files in the doc/ directory I'm not exactly sure where to start.

clean up executable installation

The spack package for fabtsuite (https://github.com/mochi-hpc/mochi-spack-packages/tree/main/packages/fabtsuite) places executables in subdirectories of the install prefix that are difficult to find, because spack packages are intentionally installed to unique locations.

It would be simpler if the commands that users should run were placed in a bin subdirectory (either in the spack package, or maybe more straightforward to do generically in the main repository).

This would also be a good time to think about the names of the executables. It might make sense to prepend a name to the commands that associates them clearly with this test suite ("fput" and "fget" are kind of generic).

how to specify which provider to test

Libfabric library builds often have support for multiple providers built in. How do you control which one is tested?

Skimming through the code I don't immediately see programmatic API control over it in the C code or environment variable control over it in the job scripts.

clarify how to specify install path

The cmake build documentation mentions two ways to alter the path (DESTDIR at make time or -DCMAKE_INSTALL_PREFIX at cmake time). See https://github.com/mercury-hpc/fabtsuite/blob/main/doc/building_cmake.md#build

Is there any functional difference between the two, or do they accomplish the same thing?

Clean up LICENSE and COPYING.

• hlog COPYING is missing. Check if https://github.com/gnuoyd/hlog/blob/main/COPYING can be used for the ECP project.
• LICENSE doesn't match https://github.com/mercury-hpc/mercury/blob/master/LICENSE.txt.

E.g.,

Copyright (c) 2021, Mercury

Add clang formatter.

Copy the formatting setting from Mercury.

clarify libfabric dependency

On the ALCF Polaris system, a Spack build of fabtsuite fails with the following error:

1 error found in build log:
     18    -- Detecting CXX compile features - done
     19    -- Found PkgConfig: /usr/bin/pkg-config (found version "0.29.2")
     20    -- Checking for module 'libfabric>=1.13'
     21    --   Requested 'libfabric >= 1.13' but version of libfabric is 1.11.
           0.4.125
     22    You may find new versions of libfabric at https://github.com/ofiwg/l
           ibfabric.git
     23    -- Could NOT find LIBFABRIC (missing: LIBFABRIC_LIBRARIES)
  >> 24    CMake Error at cmake/FindLIBFABRIC.cmake:69 (message):
     25      Could not find a suitable version of libfabric (>=1.13)
     26    Call Stack (most recent call first):
     27      CMakeLists.txt:8 (include) 

What exact features from 1.13 does the test suite require? It may be that they are actually present in this build even though it is reporting a version number less than 1.13 (the version may not track exactly in a vendor-provided library).

For reference, Mercury itself will build fine against this same version of libfabric, so hopefully it's not a particularly strict dependency.