-
Notifications
You must be signed in to change notification settings - Fork 860
WeeklyTelcon_20170808
Geoffrey Paulsen edited this page Jan 9, 2018
·
1 revision
- Dialup Info: (Do not post to public mailing list or public wiki)
- 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
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
to2048
-
rcache_base_vma_tree_items_max
to16384
-
rcache_base_vma_tree_items_inc
to2048
-
- Build issue reported to the OMPI users list:
- https://www.mail-archive.com/users@lists.open-mpi.org/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
vsppc64le
support
- 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)
- Here is one example from PR #4028
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/
============================================================================
Review Master MTT testing
- Next face-to-face meeting
- Do we need another meeting before/around Nov?
- Mellanox, Sandia, Intel
- LANL, Houston, IBM, Fujitsu
- Amazon,
- Cisco, ORNL, UTK, NVIDIA