GPFS Performance Issues

Locales: Data Centers and Campus RC Brown Research Computing
1

What’s the issue?

  • Oscar’s GPFS file system is experiencing slow I/O performance

What’s the cause?

2 things contributed to this issue:

  • GPFS was recently upgraded but some components of the file system have not yet been moved to the faster disk
  • Poor utilization of the Oscar cluster by a small number of users (thus impacting the file system for all users)

How will it be resolved?

  • GPFS metadata has been moved to the faster disk
  • The applications in GPFS runtime and the home directories are currently being moved to the faster disk
  • The Data and Scratch directories will also need to move to a faster disk
  • The small number of Oscar users who were impacting the file system had their jobs terminated and asked to work with CCV User Services

When will it be resolved?

  • Sometime within the window of July 14th-July 15th, we expect GPFS performance to be significantly better, as the migration of GPFS runtime and the home directories will be complete
  • The migration of Scratch and Data will begin the week of July 20th. More communication to follow regarding that work
2

Update on the GPFS filesystem performance: All OSCAR applications, which reside in GPFS runtime, and user home directories have been moved to a faster disk. Users should expect to see significant improvements in application startup times.

Up next: We are currently migrating the Scratch and Data volumes to faster disks and anticipate this will be completed within the next few days.

We thank you for your patience as we make these improvements. Please email support@ccv.brown.edu with any questions or concerns.

3

Update: GPFS filesystem is returned back to normal service. All filesystem-level checks indicate performance is up-to acceptable limits.