[GE dev] Modification in qsub source to match cluster policy
tvsingh at ucla.edu
Wed Mar 17 21:25:08 GMT 2010
I am mentioning this, so that other who might have situation like this, can benefit from this or developers can think of some way to incorporate similar way to address the situation. By similar I mean may be better way to address this situation.
We have a shared cluster and we run SGE as our scheduler.
The cluster is shared in the way that there are sponsor groups (people who put the resources in it) and the groups who do not. But during the time sponsors are not using their resources, the non sponsor groups could use the resources put up by sponsor group. The cluster policy also ensures that the job submitted by a sponsor will start running within 24 hrs.
After going through all the documents with in SGE, I came to conclusion that guranteeing the resources back within 24 hrs to sponsor groups are not possible. So I went ahead and deviced the following strategy:
1.Made highp complex as:
highp highp BOOL == FORCED NO FALSE 1000
2. created a queue with highp complex and sponsor access list that could use it.
3. made sure that advanced reservation is allowed.
qconf -ssconf|grep max_reservation
4. added -R n in sge_request file as default value.
5. modified qsub.c, to incorporate the following:
in case highp complex is specified in request, change "-R n" to "-R y" otherwise "-R y" by default changed to "-R n".
7. in case if sponsor submits the job with -l highp, the job gets the high priority along with request for reservation.
8. All other queues are only for 24 hrs limit, therefore if sponsor submits the job with highp complex, he sure gets the slots for his use within 24 hrs. Also non sponsor's request for -R y is not honored.
Hope this is helpful.
To unsubscribe from this discussion, e-mail: [dev-unsubscribe at gridengine.sunsource.net].
More information about the gridengine-users