The COSMA family of queues: cosma, cosma-prince and cordelia

cosma and cosma-prince

The cosma and cosma-prince queues give access to most of the the Durham COSMA4 machine, (so should really be cosma4 and cosma4-prince, but are not for historical reasons) they are for use by non-DiRAC projects (Virgo) that want to run parallel jobs (MPI/OpenMP/Hybrid), and give access to 244 compute nodes. The queues are configured so that job exclusive access is forced for jobs using 12 or more cores, but, unlike those of COSMA5, not enforced for jobs using fewer cores. To force exclusive access for smaller jobs you need to use the -x option when submitting your job.

The COSMA4 compute nodes have 12 cores and a maximum physical memory of 60Gb.

In addition to the hardware limits, the following run-times are applied:

Nameprioritymaximum run timemaximum cores
cosmaNormal72 hoursunlimited
cosma-princeHighest240 hoursunlimited

The priorities within these queues are based on fairshare, so a job from each user is considered in turn to fit into the available resources according to the queue dynamic priorities, as for the cosma5 queues. Access to the cosma-prince queue is only available on request for high-priority jobs and those that cannot run within 3 days and cannot be restarted.

cordelia

The cordelia queue gives access to the remainder of the Durham COSMA4 machine (in fact it shares a number of nodes with cosma, these are offered to cordelia first) and runs with the same limitations and queue priorities. It is for use by serial jobs that use a single core so does not offer exclusive access to any node.