Requests for queue changes should be submitted to
helpdesk@rice.edu with a copy to
odegard@rice.edu. Such requests will be collected and considered by the steering committee quarterly.
A job can request up to 4 hours wall time for an arbitrary number of CPUs.
A user may have as many as 50 active (running) jobs at a time during a period of high load, with each job running on an arbitrary number of CPUs.
A user is limited to 300 jobs total submitted to the queue.
Fair Share is implemented at a user level and insures that users' historical utilization is factored into prioritization of any new requests. It basically logs users' total CPU use over a designated time window. If all other things are equal, the less prolific of two users is assigned a higher priority via Fair Share.
The Fair Share contribution to a job's priority reaches zero when a user's consumption exceeds 10% of the total available hours from the prior week.
Scheduling is biased such that higher priority is assigned to parallel (multiprocessor) jobs.
A preventative maintenance period may be scheduled for any Tuesday between the hours of 9:00 a.m. and 1:00 p.m. The queues will be drained and the system taken off line during this period. The state of the queue and fair share quotas will be maintained and every effort will be made to avoid terminating jobs. We will notify users as soon as possible when we have determined that maintenance is necessary.
