This posting is meant as a brief recap of the RCA-SIG meetings that have occured this month. SIG-GPU Subcommittee met Sept 8. SIG-CPU Subcommittee met Sept 11. SIG-UE (User Experience) met Sept 14. 

SIG-CPU & SIG-GPU

The CPU & GPU groups discussed the need for a clearer HPC downtime plan that avoids the academic semester, but provides flexibility for DoIT to mitigate any emergencies that may arise. A proposed downtime schedule for the remainder of FY2027 can be found at this link.

Further, the CPU & GPU groups determined that their meetings are open to all members of the campus community to attend. Any elected members from either group are mentioned on the SIG webpage.

Both the CPU & GPU groups also began discussing two proposals that stand to impact chip users. Neither proposal is finalized. The October meeting of these SIG Subcommittees will ideally feature a vote on both, where any successful proposal will go to the SIG Steering Committee for final review before being deployed.

1. Slurm Priority Model

2. Data Volume Definitions and Lifecycle Management

The Slurm Priority Model represents a set of configuration changes that ensure more equitable use of the cluster resources across chip-cpu and chip-gpu. Slurm Priority only affects jobs that are waiting to run and have no access to privileged resources (i.e., 2024 hardware). No preemption rules are changed in this proposal.

The Data Volume Definitions and Lifecycle Management proposes a set of storage volumes and groups that are eligible for creation on chip: research group volumes (same as always), class volumes, award volumes, and dataset volumes. For each volume, there are specific details that tailor them to faculty member needs. E.g., read and write permissions and user/data volume review process.

Finally, the CPU & GPU groups both discussed research storage and identified no systemic issues with the Retriever Research Storage System (RRStor). The item remains on the agenda for both Subcommittees as a placeholder for discussion.

SIG-CPU

The SIG-CPU group also discussed the new “develop” partition meant as a “fast access” Slurm partition for very short-lived jobs and workflow development. The group hopes that more users will use this specialized partition. DoIT will make a specialized posting about this "develop" partition soon.

SIG-GPU

The SIG-GPU group expressed desire for a better visualization of allocated GPUs for job assignment via Slurm. The SIG-GPU group also discussed the kick-off of a new GPU Hardware Purchase Cycle and requested a presentation on usage across chip-gpu. Both of these items are being worked on by DoIT and will be discussed at the next SIG-GPU meeting.

SIG-User Experience

The SIG-User Experience group discussed new end-of-job reporting (i.e., "epilog" reporting) as well as best ways to build the group membership. All campus community members are welcome to attend!

Upcoming Meetings

The next meetings of the SIG Subcommittees are posted as events to the HPCF myUMBC Group and shared below. See the full schedule at this link.

• SIG-CPU: Oct 9, 2026 @ 1500ET in PHYS 221 or Webex (see event posting)

• SIG-GPU: Oct 6, 2026 @ 1200ET in ITE 217 or Webex (see event posting)

• SIG-Cloud: Oct 14, 2026 @ 1200ET. Webex only (see event posting)

• SIG-User Experience: Oct 14, 2026 @ 1200 in ENGR 125a or Webex (see event posting)

 Note: There is a one-time conflict with the Cloud and User Experience Meetings in light of scheduling changes around Fall Recess

May the FLOPS be with you,
Roy Prouty
Assistant Director for Research Computing & Data