WeeklyTelcon_20200609

Open MPI Weekly Telecon ---

Given the v4.1 plan (see below), we'd like to release v4.0.4 before next Tuesday.
shmat / shmdt - What is the problem scenario?
- If the sky is falling, we need to know.
- UCX, OFI MTL/BTL don't rely on these hooks.
- openib, PAMI do rely on these hooks.
- What about UCT, Portals?
- Need to communicate silent data corruption risk.
  - registration cache that mixes and matches shared memory segments and IB segements?
  - Application making sysV registration calls.
  - We need to describe to customers if you do X, Y, and Z.
    - openib assumes a working registration cache. Found an issue where the registration cache doens't work.
- How did patcher even build?
  - Looking at the code, it looks like it didn't include portions of the build that failed. :(
  - on RHEL8 with glibc 2.28-72 (CENTOS-8?)
ACTION:
- Write up of core problem case and what type processes might be affected.
- Need to know what components are affected.
- reproducer if possible (least important)

Schedule:
- Optimistic goal is to release June 30th.
- rc1 around June 22nd.
- If there are things people need to get in, please reach out to Brian and/or Jeff
Release Engineers: Brian (AWS) Jeff Squyres (Cisco)
We've come to consensus for a v4.1.0 release
- Not breaking ABI or backwards compatibility.
- Blocker moving forward is to start from the v4.0.4 tag (Tomorrow)
- NOT touching runtime!!!
- Not going to be pulling in a new PMIx version.

PMIX
- PPN scaling issue - simple algorithmic issue in this function
  - PMIX talked about it. Artem might know someone who might be interested in working on it.
  - Algorithm behind one of the interfaces doesn't scale well.
  - Not a regression. Above ~ 4K nodes, becomes quadratic.