[Chtc-users] HPC Cluster Users - Remember the "pre" partition!! Per-user core limit increased to 400!

[chtc-users@xxxxxxxxxxx]

(NOTE: This message is relevant only to those using CHTC's HPC Cluster, and not relevant for our HTC System, which uses condor)

We have two important announcements for the HPC Cluster:

1. Take advantage of the "pre" partitionÂfor shorter jobs that you want quick turnarounds for!

The "pre" (pre-emptable) partition backfills the large number of servers that are "owned" and prioritized for other researchers. Many of these servers are "idle" as some owners often have many servers idle.Â

When you submit jobs through the "pre" partition, your jobs will start up sooner than they would when waiting in the busier "univ" and "univ2" partitions. These jobs are subject to a 24-hour max time and 16-server size limit, and *may* be interrupted ("pre-empted"), but interruption is rare to-date, and pre-empted jobs will be restarted elsewhere until they finish. Read more in the HPC Cluster Guide.

To see how many servers are "idle", type the following to see ALL partitions:

sinfo -a

To see only partitions you have submission access to:

sinfo

There are currently more than 30 "owned" servers that are IDLE!! You could be using them via the "pre" partition!

2. We've increased the per-user core limit to 400!

While the "univ", "univ2", and "pre" partitions have a 16-server per-jobÂlimit, you can still have multiple jobs going in order to use up to 400 total cores across multiple jobs.Â

Previously, this per-userÂvalue was set to 320 cores. We've increased this value as a result of the major addition of hardware in the new "univ2". You may have already noticed the difference, and we may also increase this value again in the future. Furthermore, this change is now reflected in the HPC Cluster Guide.

As always, please send any questions or comments to chtc@xxxxxxxxxxx. We love hearing from you.

Happy Computing,

Your CHTC Team

(care of Lauren Michael, Research Computing Facilitator)