WeeklyTelcon_20170808

Open MPI Weekly Telcon

---

• Dialup Info: (Do not post to public mailing list or public wiki)

Attendees

• Josh Hursey (IBM)
• Jeff Squyres (Cisco)
• Artem (Mellanox)
• Edgar Gabriel (UofH)
• George Bosilca (UTK)
• Joshua Ladd (Mellanox)
• Nathan Hjelm (LANL)
• Ralph Castain (Intel)
• Nathan Hjelm (LANL)
• Todd Kordenbrock

Agenda

Review All Open Blockers

Review v2.0.x Milestones v2.0.4

• Nothing new to report.

Review v2.x Milestones v2.1.2

• PR #4042
  • Also on v3.0.x in PR #4041
  • Ref Issue #3685
  • We were not hooking madvise, but we have to after investigation.
  • VMA cache (red-black) tree implementation needs a fix as well.
    • See commits in that PR for details. Problem with concurrent memory allocations leading to deadlock, and performance.
    • Currently this adds a stop gap measure.
    • Need to redesign of red-black tree implementation...
  • New MCA parameters to control the cache:
    • rcache_base_vma_tree_items_min to 2048
    • rcache_base_vma_tree_items_max to 16384
    • rcache_base_vma_tree_items_inc to 2048
• Build issue reported to the OMPI users list:
  • https://www.mail-archive.com/[email protected]/msg31383.html
  • Jeff S. to file an issue for it so we can track it.
• RC for v2.1.2 by end of the week.

Review v3.0.x Milestones v3.0

• Atomics - builtin atomics disabled for now.
  • Goal to move to C11 atomics, but performance issues might stand in our way.
• XLC related tickets
  • https://github.com/open-mpi/ompi/issues/3816 (Close as won't fix, compiler too old)
  • https://github.com/open-mpi/ompi/issues/3817
  • https://github.com/open-mpi/ompi/issues/3811
  • Josh to work on a configure check for XLC version, and ppc64 vs ppc64le support
    • https://github.com/open-mpi/ompi/issues/4053
• Hostfile behavior change (https://github.com/open-mpi/ompi/issues/3984)
  • Ralph listed as a non-blocker since it changes behavior.
• A number of blockers still outstanding that need to be addressed
• Want an rc -soon- but probably not until Monday at the earliest.

Review Master Master Pull Requests

• make check failure with opal_fifo/opal_lifo issue found by CI tests
  • Here is one example from PR #4028
    • https://jenkins.open-mpi.org/jenkins/job/open-mpi.pull_request/989/
    • Suggest getting more information from the make check (add serial-test to automake)

make[4]: Entering directory '/PATHTOTESTDIR/src/test/class'
PASS: opal_bitmap
FAIL: opal_hash_table
FAIL: opal_proc_table
FAIL: opal_tree
FAIL: opal_list
FAIL: opal_value_array
PASS: opal_pointer_array
FAIL: opal_lifo
FAIL: opal_fifo
============================================================================
Testsuite summary for Open MPI Run Time Environment gitclone
============================================================================
# TOTAL: 9
# PASS:  2
# SKIP:  0
# XFAIL: 0
# FAIL:  7
# XPASS: 0
# ERROR: 0
============================================================================
See test/class/test-suite.log
Please report to http://www.open-mpi.org/community/help/
============================================================================

---

MTT / Jenkins Testing

Review Master MTT testing

MTT Dev status:

Jenkins CI

---

Exceptional topics

• Next face-to-face meeting
  • Do we need another meeting before/around Nov?

---

Status Updates:

Status Update Rotation

1. Mellanox, Sandia, Intel
2. LANL, Houston, IBM, Fujitsu
3. Amazon,
4. Cisco, ORNL, UTK, NVIDIA

---

Back to 2017 WeeklyTelcon-2017