Conference call - May 18th - 19:00 UTC
Dear all, the next DRMAA conf call is scheduled for May 18th, 19:00 UTC. We meet on Skype, please find me under my user name "potsdam_pit". Preliminary meeting agenda: 1. Meeting secretary for this meeting? 2. Review of advance reservation research work by Mariusz (http://bit.ly/kLwNe7) 3. Discussion of usersACL proposal by Mariusz 4. Discussion of NOW constant proposal Best regards, Peter.
Participants: Roger, Daniel G. , Andre (SAGA), Mariusz
2. Review of advance reservation research work by Mariusz (http://bit.ly/kLwNe7)
- MOAB not part of the analysis, but only commercial version of MAUI - Daniel G. approved analysis result for SGE - Available and understood: reservationName, startTime, endTime, minSlots, maxSlots - machineOS, machineArch: Supported in all systems, but still demand from group to make it optional - INFINITY in endTime not broadly covered, to be removed - candidateHosts broadly supported - half of the systems does not support automated choice of a subset of the given machine list - only all-or-nothing semantic can be promised - subset semantic to be added as MAY feature of an implementation - minPhysMemory - Difference between demand for according node, and true reservation of given memory amount - Implementation detail from DRMAA perspective - ReservationInfo attributes are fine, reservationOwner not broadly supported -> remove - SLURM and LoadLeveler support only node, not slot reservation -> proposal to report 'reservation granularity' as capability
3. Discussion of usersACL proposal by Mariusz
- Defines list of users that are allowed to use the AR - SAGA has notion of permissions and user ID's, can live with / without it - Andre: Another typical use case for AR is co-scheduling - Problem how to transport AR ID from creator application to consumer application (mismatch with DRMAA session concept) - In out-of-the-box installation, standard users cannot create AR's - Hot discussion without agreement, deferred to public poll on the mailing list
4. Discussion of NOW constant proposal
- According to Mariusz investigation, only partial support for NOW in startTime -> to be optional
participants (1)
-
Peter Tröger