Case study · A university supercomputing facility
Checkpoints that don’t fill the cluster.
Cutting simulation checkpoint storage and write time on a shared HPC system
- HPC centres
- Research computing
- Simulation teams
Download the case study8 pages · PDF · 2026
- smaller checkpoints
- 10×
- smaller checkpoints
- faster checkpoint writes
- 3×
- faster checkpoint writes
- more runs kept online
- 4×
- more runs kept online
What’s inside
The whole story, with the numbers.
The cluster
Thousands of nodes, a shared parallel file system and checkpoints that crowded out everything else.
Faster writes
Compressing before the write, so jobs spend less time stalled on I/O.
On-premises install
Running inside the facility’s own network, with no data leaving it.
Rolling it out
Bringing research groups across one at a time, starting with the biggest users.
Contents
4 chapters.
- 01Storage pressure on a shared system
- 02Lossless where it matters, bounded where it doesn’t
- 03Installing on-premises
- 04Results across research groups
Free download
Where should we send it?
Fill this in and the download link appears straight away.
It’s yours.
Rather see it on your own data?
Every demo is run by the people who build Compress, on a sample of your own files.