Please make reservations for cluster capacity if you plan to submit very large jobs that significantly exceed the current maximum "free" allocation.
Please send reservation requests to email@example.com at least 24 hours before the reservation is required to go in effect. Note that a reservation request can only be satisfied if the required resources are available during the requested period and this cannot be guarateed until the reservation is actually created.
Reservations will be created on a first-come, first-serve basis.
Some reservations will be created with a "byname" flag. As a result, you will need to use "-l advres=<reservationID>" (this ID will be sent to the the requestor via email) as a part of your batch script to actually use the reservation.
If at any point in the reservation duration, the utilization of the reservation falls below 50% for a period of 2 hours, then the reservation will be cancelled to make these resources available to other jobs/users. Remaining jobs will continue running. For example if for 30 CPU reservation submitted jobs only use less than 15 CPUs for 2 hours, the reservation will be canceled.
Up to half of the cluster will be dedicated to reservations.
If you want to submit a group of jobs requiring 30 CPUs for 30 days, your effective walltime requirement is 900 days. But the current cluster-wide default walltime is 100 days and so some of your jobs will be queued instead. If you want all the jobs to run right away, you could request a reservation for the resources by emailing firstname.lastname@example.org.
Once a reservation is created, you will be emailed a Reservation ID (for example, user.100). Please submit jobs using the following syntax in case of command-line options.
qsub -l advres=user.100 <other flags>
In case of a batch file submission, please add the following to your batch file along with other PBS directives.
#PBS -l advres=user.100
One-time reservations (created on user request):
|Reservation ID||User||Nodes||Processors||Starting Date / Time||Duration|