The two shared memory queues give access to the COSMA5 and COSMA5 login nodes, so share with users submitting jobs, developing code, doing post run analyses etc. They are intended for use by jobs that require a large amount of memory (greater than that of the compute nodes) and for no other reasons, except small amounts of testing.
The shared memory nodes have a mix of available cores and memory limits, as well as architectures:
|Name||queues||maximum cores||physical memory||architecture|
COSMA4 users can submit to the shm4 queue and COSMA5 users to
shm5 using the
projects as for the cosma and cosma5 queues.
Clearly, given this mix, it may be important to control which nodes or architectures you need for your job, that is explained in the submitting jobs to specific architectures description.
Neither queue has any run-time limits, but, since these are intended for shared memory jobs and share with interactive users, before a job will run a node must have at least 50% of its physical memory free and have a current load that is on average only using half the number of cores (the actual numbers of cores are roughly twice the numbers available to the queues). It is also necessary to specify an upper memory limit for the job, if it exceeds this value the job will be killed.