Is there any way of restricting the total amount of disk space used per user? And related, what happens when a job fails with an “No space left on device” error? I mean this for the shared file system, not the local SSD cache.
David
Hi @DavidHoover,
Unfortunately, this isn’t possible. It might be possible by using OS-level tricks (partitioning a disk?).
This means that a job that is writing out results (e.g. micrographs for a Patch Motion job or particles for an Extract From Micrographs job) tried to write a file to the disk, but failed because there wasn’t any space. The job will fail, but you can clear it and re-run it once the volume is cleared up.
1 Like
If the job fails and there is no database corruption, then no problem. How does CryoSPARC handle NFS? Does it ever attempt I/O intensive tasks?
That’s correct, it’s harmless. Though I do suspect there will be an issue if the disk the database is hosted on runs out of space; it’s best to keep this on a separate volume than the project directories.
At Structura, we use NFS to mount large storage volumes onto each one of our GPU workstations. We have a 10GB link between the nodes, and we get solid preprocessing performance. Intensive I/O tasks (e.g. random reads required for particle processing in jobs like Non-Uniform Refinement) are reserved for the SSD cache.
Does Structura use NFS v3 or v4?
Our mounts are using v4.2.