Skip to content

2017 04 26

Jump to bottom
Wesley Bland edited this page May 9, 2017 · 2 revisions

Attendees

  • Intel - Wesley, Marc, Rob
  • Argonne - Yanfei, Ken
  • LLNL - Ignacio
  • Auburn - Nawrin
  • UTK - Aurelein

Slides

In repository

Reinit

Ignacio presented an overview of the Reinit proposal that has more details than the header file we saw a few years ago.

There were a couple of questions:

  • Wesley - Are processes that are RESTARTED guaranteed to have the same rank?
    • Yes
  • Rob - Is this compatible with Fortran?
    • It uses a longjump internally and user-facing function pointers. We'd have to think about how to tweak that to support Fortran.
  • Yanfei - So this relies on the resilience of Slurm (or any process manager)?
    • At the moment, that is true. From a specification perspective, it could all be implemented inside MPI. From a practical perspective, it's nice if the process manager can take care of it since many already do some of that work.
Clone this wiki locally