Visualizing resources managed by a PBS job scheduler

Aug 31, 2019

At the research institute I currently work at, we use PBS as a job scheduler for collaborative use and research management of the HPC cluster. Just for a little background info, the cluster consists of 14 computers which are connected to make the cluster. Two of those 14 are masters, one of them being a backup of the first. The master isn’t supposed to be used to for computation because its main job is to distribute the work between the other machines. The other 12 computers, also called nodes, are those which we do all our work on. The job scheduler runs on the master and assigns jobs to the nodes. A job is a set of operation you need to be done and which is saved in a script which you hand over to the job scheduler with the request to execute it. In bioinformatics, a good example would be the alignment of reads to a reference genome. As mentioned before, we use PBS but there are others out there, e.g. SGE or Slurm. Which node gets to do which job depends on the resources (number of CPUs and memory) each job requests. Ideally, this allows efficient resource usage so that everybody gets their job done the quickest.