Low Priority Queue allows condo contributors to glean unused cycles on Savio

Architectural overview, Savio cluster

The Berkeley Research Computing Program now offers Savio users who have contributed nodes to the condo cluster a way to run more and/or larger jobs at no additional cost.

A new “Low-Priority Queue” allows condo contributors to schedule additional jobs that utilize unused capacity on nodes used by those whose access is granted under the Faculty Computing Allowance. These jobs may exceed the size of (i.e., the number of nodes in) the contributor’s condo, with per-job limits capped at 24 nodes and 3 days of runtime. Additionally, these jobs can be run on all types and generations of hardware resources in the Savio cluster, not just those matching the condo contribution. At this time, there is no limit or allowance applied to this QoS, so condo contributors are free to use as much of the resource as they need.

In order to ensure that other users are not impacted (e.g., users running jobs under the Faculty Computing Allowance), jobs running in the “Low Priority Queue” are subject to preemption (i.e., termination) when general queues, such as “savio_normal” and “savio_debug,” become busy. The BRC team is monitoring impact of this new queue on overall cluster utilization and performance, and the availability and/or scope of this Low-Priority Queue service may be adjusted over time as a result of actual impact.

This new offering is an additional incentive for groups to invest in condo nodes on Savio, while maximizing the campus investment in the cluster and maintaining quality of access to the shared resource for researchers using the Faculty Computing Allowance.

E-mail the BRC HPC team at brc-hpc-help@berkeley.edu if you have any questions about the new “Low-Priority Queue” service.