I am using group quotas in my setup, in the following manner:
<![if !supportLists]>· <![endif]>I have three groups, and I allocate 0.75 of the cluster's slots to them. Let's say each group gets a quota of 0.25.
<![if !supportLists]>· <![endif]>The rest I leave unallocated.
<![if !supportLists]>· <![endif]>I have configured PREEMPTION_REQUIREMENTS to allow preemption if the submitter is under its quota and the running job belongs to a group which is over the quota. So, each group is guaranteed its quota, usually without waiting.
My intention is that empty slots can be used by any group, even if the group is over its quota. From the documentation I learned that I can use both GROUP_AUTOREGROUP and GROUP_ACCEPT_SURPLUS for this purpose.
Can anyone explain the difference between the two parameters?