The HPC will be upgraded during the period of August 17 until Augustus 28. During this period we will re-stack the compute nodes from the 2007 acquisition, reconfigure part of the interconnection network and add 128 new compute nodes. The HPC will be unavailable to users during these two weeks.
The schedule for the upgrade will be:
- Sunday August 16th, evening: the job scheduler will no longer accept new jobs.
- Monday August 17th, 9AM: access to login nodes will be turned off and compute nodes will be turned off.
- Friday August 28th, afternoon: access to the cluster will be turned on again. If the upgrade goes faster than anticipated, we might open earlier and announce this to the users.
After the upgrade, the HPC cluster will consist of two subclusters:
Pictures of the upgrade from Monday morning 8/17:
- a cluster with 512 cores with its own dedicated infiniband data network. This cluster is reserved for backfill jobs and general access jobs; every FSU affiliated user can submit jobs to the general access queue.
- a cluster with 2048 cores connected to the existing infiniband network. This cluster is mainly for units that bought in, but a small part is also reserved for general access jobs.
- Both clusters will have access to our distributed parallel filesystem.
- The whole HPC system will have 5 TB of memory (2 GB per core).