NERSCPowering Scientific Discovery Since 1974

NUG Webinar April 11, 2013

Agenda

  • NUG 2014 Planning: NERSC 40th Anniversary
  • NERSC Achievement Awards 2013
  • Edison-forum mailing list
  • Batch wait time statistics
  • Edison update: Timeline for Phase 2
  • Miscellaneous
  • Special Discussion Topic: Fair Share Scheduling on Edison

 Meeting Summary

 Attending: 36 NERSC users 

  • NUG 2014 Planning: Frank Tsung (UCLA) will coordinate finding volunteers from NUG to help NERSC plan the event, which will coincide with NERSC's 40th anniversary. The meeting will likely be in Berkeley / Oakland, but the dates and agenda are still TBD. It was agreed that sometime in February would be good. Gary Grest (Sandia, BES) has also volunteered to serve on the committee. (Thanks Frank and Gary!)
  • NERSC Achievements Awards for 2013 (to be announced at NUG 2014). A committee of six from NUGEX will work with NERSC to define the nomination and selection process and to collect and review nominees. The committee will also sollicit outside reviewers. Stephane Ethier (Princeton PPL, FES) and Cameron Geddes (Berkeley Lab, HEP) volunteered to serve on the committe. (Thanks!) We are still seeking representatives from BES, NP, ASCR, and BER.
  • Edison email list: Many people signed up, but few are posting. Users said they would prefer NERSC staff to post more often and that would start conversations on various topics. 
  •  Dave Goodwin reconfirmed that usage on Edison is not expected to be charged against repo allocations through CY 2013.
  • There was a request to have a place in /usr/common for groups to maintain software and modules. It was suggested that this could be magaged like /project, with groups being given their own subdirectories with group permissions. File access time from /project were not deemed adequate for this purpose.
  • NERSC is experimenting with a fair share scheduler on Edison (see presentation slides). There was concern that this would prevent users from getting good throughput at certain times that are important to them (e.g. important conferences). Others pointed out that fair share might actually benefit those in this situation if the "decay rate" was set long enough. There were questions about how high-throughput workflows would work under fair share. NERSC will be monitoring and users should express their opions to NERSC consultants.
  • A NUG working group to advise NERSC on queue policy is forming. Anubhav Jain (Berkeley Lab, BES) and Stephen Bailey (Berkely Lab, HEP)) volunteered to be members. We're looking for representatives from other science areas and with varied workflow needs.
  • There is a need to run applications like matlab, IDL, etc. They should not be run on login nodes if they are compute intensive. NERSC promised to better advertise and document how to use the large-memory nodes attached to Carver.

Please send feedback and suggestions to consult@nersc.gov or Richard Gerber (ragerber@nersc.gov)

Richard Gerber

Downloads