> Most options, like this one, require trusting the user to play > along. If they don't/won't. Then you can enable preemption and setup > policy to de-prioritize the user's jobs the more they have > concurrently running. This is like running condor_vacate_job on > their jobs. > > Ian can probably give you a handful of configs that do that too. Getting preemption or suspension would also work, is there a configuration that also tries to limit the number of times a job gets restarted? While I had pre-emption enabled, I had an occasional failure mode where a job would keep getting assigned to a system with too little ram and never actually finish. Diane
Attachment:
signature.asc
Description: Digital signature