Skip to content

Meeting 2020 02 14

Stephen Herbein edited this page Feb 14, 2020 · 1 revision

Agenda

Notes

Joshua Hursey (IBM)
Michael Karo (Altair)
Ralph Castain (Intel)
Kathryn Mohror (LLNL)
David Solt (IBM)
Shane Snyder (ANL)
Stephen Herbein (LLNL)
Swaroop Pophale (ORNL)

Stephen’s Notes:

  • WG update: Slices/Grouping of functionality
    • Continuing to make progress on integrating feedback from ASC quarterly into use-cases
    • Thinking about doing a tech talk at the ASC about modex and other nuances of business card exchange
  • WG update: Storage
    • Focused on event notification system and system query interface to see how they can integrate storage functionality into those - e.g., query topology/characteristics of the storage system
    • Collecting feedback on what users want in term of storage functionality that would be new to the standard - e.g., data movement
    • Josh: have you thought about the prototype vector that you’d like to target?
      • Shane: lustre would be a good starting point and easy to get access to
  • WG update: Dynamic Workflows
    • Switched to biweekly
    • Discussion and level setting of what we mean by dynamic workflows/jobs - e.g., spark, kubernetes
    • Talk from radical pilot folks
      • Thinking about how they might use PMIx_Spawn vs the native research manager
      • Slides posted to the WG mailing list
  • WG update: implementation agnostic document
    • Posted an update to Chap 1 ticket - straw poll to continue on with ticket readings and to consider changes not warranting of a reading reset
    • Chapter 2 PR posted with current updates to current thoughts/directions
    • Ralph: how do we want to handle merging these PRs in while minimizing conflicts with V4 changes
      • We will handle that issue further down the road, but conflicts with Chapters 2+ will be more than with Chapter 1
    • Presentation from Dave on Required vs Required Attributes
      • Link to slides: TODO
      • Stephen: do you have a suggestion on how to differentitate attribute not supported vs value of the attribute not supported
      • Ralph: there was a proposal for how to provide an error status for each individual attribute (increasing granularity of the errors)
      • Dave: could we extend the status return to add information about whether the error was in the attribute or the value
      • Ralph: we can use the flag field in the status struct, but we need to go back through a look where const is set
      • Ralph: may make sense to build a matrix of all the possible scenarios and make sure they are all covered, think through “what happens and what did we want to happen”
      • Ralph: I’ll take the action item to make a first pass on these questions and present to the two working groups
Clone this wiki locally