.. index:: calculations .. _running calculations: Running calculations ==================== Most workstations are not in heavy use all the time. To maximise our resources, the workstations are set up to run calculations via a job submission system, much in the same way many compute clusters operate. See :ref:`slurm` for details on how to use this system. Please submit jobs via Slurm when running long and/or several concurrent calculations. The maximum walltime is 5 days. This is to prevent too much wasted time in the case of power outages, reboots, etc. and to (hopefully!) encourage the writing of programs that can be restarted, which is a good practice both to achieve throughput on larger compute clusters and to avoid wasting large amounts of computer time if a computer does crash. .. important:: Some calculations can produce large volumes of output. As noted in :ref:`home`, extremely heavy i/o on the NFS server can freeze the cluster. This is particularly likely if many calculations are running at the same time. Such calculations will be killed without warning. It is important to instead run calculations in your :ref:`workspace` directory and (if needed) copy files to your home directory once the calculation has finished. .. note:: Most workstations are quad-core processors with 16GB RAM. Please remember that someone is using each workstation as their computer and do not submit calculations which will use excessive amounts of memory, as this will cause the performance for the local user to be unacceptable. Such jobs are liable to be killed without warning. Currently Slurm calculations are restricted to using a single core and 4GB RAM. It is advisable to prefix computationally-demanding commands in your submit script with 'nice -n 19' to prevent your calculation taking up too many resources. Nevertheless, it is possible to get some useful calculations done on the cluster, depending upon the required resources. For larger calculations (either in terms of throughput, or memory use or CPU use) please see the College's `High Performance Computing `_ service; in particular the HPC service wiki (linked to the from HPC website) contains a lot of useful information.