Pegasus Summer Maintenance Notes
We are excited to offer a “heightened memory CPU node” configuration with 384 GB; to distinguish these from our “high memory nodes” (3 TB) we refer to this configuration as “medium memory.” This will better accommodate increased requests for applications requiring more than a 192 GB/node. We have gathered memory utilization data for several months, and over 75% of CPU jobs used 96 GB or less. The “default” CPU node configuration will be 96 GB, allowing 54 nodes of the “heightened memory” configuration 384 GB. This will allow for GPU-enabled nodes to be restricted to GPU jobs.
Over the last several months, as Pegasus became increasingly busy, we observed that shorter runtime jobs waiting several hours to dequeue created a poor user experience. To improve cluster responsiveness, there is now a “nano” partition to serve jobs that are expected to complete in less than 30 minutes.
During this maintenance period we successfully:
- Created a 54-node, 384 GB “384gb” Memory partition
- Created a 30-node “nano” partition
- Implemented a policy requiring explicit specification of the partition on which to run, i.e., your submit script must now include the line
- #SBATCH -p partition_name #or
- #SBATCH --partition partition_name
- Included eleven 384 GB nodes in the short partition and created a partition named short-384gb as a subset.
We thank you for your patience as we performed updates. If you have any questions please email hpchelp@gwu.edu.
// To keep up with upcoming maintenance windows and other HPC related events subscribe to the HPC Support calendar here or view on our website.
Thanks,
- GW HPC Team
Posted June 11, 2021
