Conference call - Apr 14th - 19:00 UTC
Dear all, sorry for the late announcement ... The next DRMAA phone conference is scheduled on Apr 14th, at 19:00 UTC. For time zone conversion, check this link: http://timeanddate.com/s/1pn9 The phone conference line is sponsored by Oracle: Phone number (toll-free from US): +001-866-545-5227 Access code: 5988285 Preliminary meeting agenda: 1. Meeting secretary for this meeting? 2. Queue discussion (relationship to slot concept, relationship to machine concept) 3. Parallel environment discussion (machine list availability, configuration name semantics) 4. Support for core file fetching 5. Support for marking re-queueable jobs 6. JobTemplate structuring proposals Best regards, Peter.
Participants: Roger, Daniel, Peter
1. Meeting secretary for this meeting?
Peter
2. Queue discussion (relationship to slot concept, relationship to machine concept)
- New slot concept is important on its own, regardless of queue relationship - DRMAA cannot assign any semantic to the queue term - could be 'waiting line' (non-SGE) or 'execution host' (SGE) - SGE can have one job in multiple queues - Practice shows that site admins sometimes describe queue semantic in long queue names - Stick with 'list of opaque queue names' approach - From portal application view point, queueMaxSlotsAllowed is as valid as queue name listing - Implies no meaning for what a queue is, gives only maximum allowed value for a job template parameter - Daniel started to collect other possible queue monitoring attributes on Google Docs page - Getting information about queue - machine relationship is only useful for debugging purposes, nothing for DRMAA - Requesting a number of slots for a job in one queue has no implication on number of utilized machines - MPI vs. OpenMP case only depends on configuration aspects - Question about relationship between user and monitoring list information (e.g. list of queue names) - Can be system view (qmon case) or user-centric view -> which queues are available for me - Decision for user-based perspective, also based on practical experience in the field (Roger) - Demands queue ACL checks in the DRMAA implementation - Should be general approach for MonitoringSession interface
3. Parallel environment discussion (machine list availability, configuration name semantics)
- Configuration name proposal page looks reasonable - Mapping execution environment conditions is clearly an installation / site-specific issue - DRMAA should only provide name proposals and their meanings
4. Support for core file fetching
- Basic assumption that core file ends up in the working directory - Problem: Staging is over when the JobInfo information is available - Level of DRM system support for core file fetching is unclear Best, Peter.
participants (1)
-
Peter Tröger