SLURM: Half load on hyperthreads
I spent more time than I care to admit this afternoon trying to understand why SLURM was underutilising CPUs on my compute node. Worse still, I had encountered and resolved the same problem several months ago, so I am putting it here in case it is useful to someone else (i.e. me, in another six months1). Behaviour ~50% system load, despite full CPU allocation according to scontrol2: $ scontrol show node mynode NodeName=mynode Sockets=2 CoresPerSocket=40 ThreadsPerCore=2 CPUAlloc=160 CPUEfctv=160 CPUTot=160 State=ALLOCATED CfgTRES=cpu=160 AllocTRES=cpu=160 Half load with full allocation smells of a misunderstanding between physical and logical cores....