Hi,
I'd like to ask a question about how the disk usage ad gets used. I see that every node shows the Disk ad, which shows how much space is left in the partition the condor work area is.
You can add the request_disk attribute to the submit file, and that will allow the job to be brokered to a node which has enough disk space, but my question is does this take into account all the request_disk requests currently on the node, or just the value of the Disk ad?
e.g.
You schedule one job, which requested 40GB of disk space via the request_disk attribute, but it will only ever use 3 GB. It got scheduled to node A, which has 70GB of total space for jobs. Another job comes in and requests 40GB of disk space too. Can the second job run on the same node, because the Disk ad will always be > 40GB? Or does the scheduler know that the currently running job requested 40GB, and therefore could use that space in the future, and therefore won't schedule the other 40GB job on there too?
I'm trying to solve a problem on making sure my nodes never run out of disk space, and don't know if I should add request_disk to the jobs that come in, or use a new machine resource.
Cheers,
Sean
--
Sean Crosby
Research Computing | CoEPPÂ| School of Physics
Senior System Administrator |ÂHPC | Research Platform Services
University of Melbourne
![[Computer Systems Lab]](https://www-auth.cs.wisc.edu/pics/csl_logo.gif){kind=logo}