Our Cray XCs have been running with a node defintion of
Gres=craynetwork:4 in the config for a good while now, even though we have not activated the "use" of the Gres via the JobSubmitPlugins=job_submit/cray_aries setting. On our TDS, however, the JobSubmitPlugin was active, and the upgrade to 20.11.x, and the use of --exact in multi-step jobs saw us comtemplating the need to tell our user community that they would need to specify --gres=craynetwork:0 in such jobs to stop the first srun grabbing all 4! Thankfully, if we don't activate the plugin. then our user community don't need to do anything, and maybe don't even need to know about the craynetwork GRES. History shows that the various craynetwork-related settings, both in slurm.conf and gres.conf, come from Cray's add-on utility, slurmconfgen_smw.py which was used to generate our "seed config" file, way back when. I'd like to remove all references to the craynetwork GRES, in slurm.conf and gres.conf as it doesn't seem to do anything. My initial testing suggest that this will be OK, but ... has anyone out there operating a Cray XC, done this, or done this and found they had to put some craynetwork bits back in to keep their Crays happy? Kevin -- Supercomputing Systems Administrator Pawsey Supercomputing Centre
