In my last role we moved from SGE to Slurm.
However we did this by using VMs for all the control, login, slurmDBD and MariaDB nodes, so it was easy enough to build a Slurm cluster up to the point where it needed compute nodes. We then removed compute nodes in groups from SGE, reinstalled with CentOS 7 and then added them to the Slurm cluster. Previously they had been CentOS 6 but that was unsustainable. Doing it this way removed the time pressure as we could allow users to use both in parallel for a while, and so it was not so hard to switch off the SGE cluster. We also started using Lmod (and EasyBuild) which hugely improved the management of installed software, but that is an aside. The only issues that we had were where we had some locations in $PATH that still had SGE binaries like qsub. We installed the slurm-torque RPM as it provides wrappers that allow us of familiar commands like qsub with Slurm. We just had to make sure that the Slurm wrappers were found first in the $PATH. You may have to work out where those are as I expect that you want to keep the current OS build if you are suing xCAT. There is then the small matter of retraining users; there is a ‘rosetta’ PDF out on the Internet which shows equivalent commands for several schedulers, which was helpful. Most users were not expert in SGE so did not have complex scripts or a great desire to keep it. There were a few bioinformatics packages that support SGE and do not really support Slurm but they are few. There is a lot more support (like here) for Slurm than I think that you will find for SGE. William From: slurm-users <slurm-users-boun...@lists.schedmd.com> On Behalf Of Nguyen Dai Quy Sent: 12 November 2019 14:38 To: Slurm User Community List <slurm-users@lists.schedmd.com> Subject: [slurm-users] Replace SGE by Slurm on running cluster Hi list, We have a small HPC Linux Cluster (CentOS 7, xCAT,...) with 8 nodes running actually with SGE. We would like to replace SGE by Slurm. Do you have any experience with this kind of work? Thank you,
