Date:
Tuesday, May 23, 2017 - 8:00pm
System(s):
Cluster Computing
- Batch system changes
- Add 'pfsdir' as a node attribute to allow creation/removal of PFSDIR for jobs that request the attribute. New location of PFSDIR is /fs/scratch/USER/JOBID. JOBID includes the job number, and batch server name.
- Update submit filter to be Python 3 safe.
- Reorder Oakley batch queue routing so jobs for the large memory nodes are put into the correct queue.
- Restructure GPU reservation on Ruby to segregate 'vis' jobs onto dedicated nodes
- License server changes
- license[1,4] servers turned off.
- Legacy Intel license server on license2 is no longer available (statewide users were to migrate to license5 long ago.)
- Partek license server on license2 is no longer available
Research Data Storage
- Project & Scratch
- Perform file system check ('mmfsck') on both project, and scratch file systems
- Enable DMAPI support for future use with IBM SOBAR backup scheme, and eventual user accessible archive to tape
- Change tuning parameters per recommendations from DDN
- Upgrade GPFS to 4.2.2-3 on protocol, and IME servers
- Home directories
- Enable >16 group support for /users/ NFS mounts.
- Still do not have support for >16 groups on /, /usr/local/, etc. This is due to a problem with setuid failing that could not be solved before the downtime.