Hi; We're using Slurm to manage jobs that require the use of licences. These licences are by far the limiting resource rather than hardware (they are very expensive; hardware is relatively cheap). We would like to have two partitions - a 'low priority' partition and a 'high priority' partition. Jobs submitted to the high priority partition should pre-empt jobs in the low priority partition. In being pre-empted, a low priority job should surrender its licences back to the pool to be used by the higher priority job.
I understand that Slurm isn't capable of this out-of-the-box, so have been making some modifications to achieve what we need. Thus far I've modified slurm to release licences when jobs are suspended, and reclaim them when a job is resumed. This works well when jobs are manually suspended and resumed using scontrol, but I'm a little lost on how to modify the scheduler/pre-emption algorithm to realise that it can release licences by suspending jobs. I suspect I need to change some things is src/plugins/select/select_cons_res.c - is this correct? Some light googling suggests that this question has been asked before, but the email train doesn't seem to go very far: https://groups.google.com/forum/#!topic/slurm-devel/urUiy53WGqA Any pointers would be appreciated. Kind Regards Fugro AG Pty Ltd Sam Ingarfield Geotechnical Engineer T: +61 8 9423 3300 | D: +61 8 9423 3317 [email protected]<mailto:[email protected]> | www.fugro.com<http://www.fugro.com> Level 2, 52 - 54 Monash Ave, Nedlands, WA, 6009, Australia A.B.N: 25 064 487 298
