Thurs 9/3 telecon

4 views
Skip to first unread message

Ralph Castain

unread,
Aug 31, 2026, 4:05:35 PMAug 31
to pm...@googlegroups.com
Hello all

Hard to believe it is already Sept!! We will have our usual monthly teleconference at the usual time (11am US Eastern) and place (https://openpmix.org/captcha/).

The summer of action has continued, but is close to completion. I’ve spent a ton of time working with Claude on a deep review of the entire PMIx and PRRTE code bases, and working with Petter on adding elasticity to PRRTE. Happy to report a lot of progress on both counts.

PMIx activity:

* complete revamp of the shared memory system to support collection of fence-passed data (what we call “module exchange” and others call “business card”), significantly enhance performance by removing all locks from the PMIx_Get operation, support modification/deletion of keys, and elastic operations

* construction of an extensive test harness that includes 100s of unit tests and a large number of Docker swarm tests for multi-node operations. Every bug repair and enhancement has been paired with a unit test for correctness and regression detection

* addition of a couple of APIs for supporting registration/deregistration of sessions to support dynamic environments

* cleanup of a wealth of error paths and race conditions


PRRTE activity:

* significant performance improvements in the grpcomm collectives as we streamlined the data being sent and the patterns being used

* construction of an extensive test harness that includes 100s of unit tests and more than 1000 Docker swarm tests for multi-node operations, including a Slurm swarm for checking that environment. Every bug repair and enhancement has been paired with a unit test and/or swarm script for correctness and regression detection. NOTE: running the full swarm test suite takes just over 2 hours due to the number of tests plus node and radix combinations

* continued work on elasticity support, particularly in the Slurm and PMIx-scheduler arenas

* cleanup of a wealth of error paths and race conditions


Look forward to seeing you there and discussing these things!
Ralph

Reply all
Reply to author
Forward
0 new messages