[drmaa-wg] DRMAA WG minutes Aug. 9th 2005

[Andreas Haas]

Minutes:

Hrabri Rajic
Peter Troeger
Roger Brobst
Daniel Templeton
Andreas Haas

Minutes from last meeting
   Accepted.

Minutes from GGF14 (see below)
   Talked about GGF14.
   Accepted.

How to continue with DRMAA?
   * in order for DRMAA to become GGF recommendation we need to
     continue with GGF document process
   * next milestone is GGF.E document
   * thus further DRMAA WG activities and discussions should be
     done with GGF.E document in mind as *the* target

Discussion
   Q: Do we need DRMAA_ERRNO_UNSUPPORTED_ATTRIBUTE be applied for
      introspective bindings?
   A: It should be allowed but not be a must.

 -----------------------------------------------------------------

GGF14 DRMAA Session Notes

Presentation by Peter (see slides)
Question (Bob Namestkan from SUN): How can I access my jobs when the session crashes,
in case the application persists the job identifiers
  Solution 1: Notion of unique session ID, which allows a re-opening of a session
              Would demand persistency features in a DRMAA library, which is bad
  Solution 2: Allow to ask for status of jobs, which are not in the current (newly created) session
              Would move the problem completely to the user space, which is good
              Would give a different semantic to wait(SESSION_ALL), which is bad
              Demands the promise of a globally unique job identifier

COG on-top-of DRMAA is a great idea, since COG offers unique workflow features
Positioning of DRMAA vs. SAGA is unclear, SAGA seems to be a replacement
Michihiro Koide, IBM Japana Systems Engineering: Working on-top-of Globus,
complicated API, searches for a good alternative, will offer ideas for improved
DRMAA spec
Andre Merky (SAGA): SAGA on-top-of DRMAA would be great to have
Suggestion: DRMAA on-top-of SSH
Lot of interest in the DRMAA GridWay implementation