The man page for slurm.conf, select/cons_res parameter SelectTypeParameters, values CR_Socket and CR_Socket_Memory states the following:
"Note that jobs requesting one CPU will only be given access to that one CPU" I think this statement is incorrect, or at least very misleading to users. A job requesting one CPU will only be allocated one CPU, but unless task/affinity is enabled or some other CPU binding mechanism is used, the job can access all of the CPUs on the node. That is, a task that is distributed to the node can run on any of the CPUs on the node, not just on the one CPU that was allocated to its job. I propose the following patch to replace "given access to" with "allocated". Regards, Martin Index: doc/man/man5/slurm.conf.5 =================================================================== RCS file: /cvsroot/slurm/slurm/doc/man/man5/slurm.conf.5,v retrieving revision 1.1.1.50.2.1 diff -u -r1.1.1.50.2.1 slurm.conf.5 --- doc/man/man5/slurm.conf.5 8 Mar 2011 13:39:41 -0000 1.1.1.50.2.1 +++ doc/man/man5/slurm.conf.5 29 Mar 2011 20:28:53 -0000 @@ -1495,7 +1495,7 @@ On nodes with multiple cores, each core or thread is counted as a CPU to satisfy a job's resource requirement, but multiple jobs are not allocated resources on the same socket. -Note that jobs requesting one CPU will only be given access to +Note that jobs requesting one CPU will only be allocated that one CPU, but no other job will share the socket. .TP \fBCR_Socket_Memory\fR @@ -1503,7 +1503,7 @@ On nodes with multiple cores, each core or thread is counted as a CPU to satisfy a job's resource requirement, but multiple jobs are not allocated resources on the same socket. -Note that jobs requesting one CPU will only be given access to +Note that jobs requesting one CPU will only be allocated that one CPU, but no other job will share the socket. Setting a value for \fBDefMemPerCPU\fR is strongly recommended. .TP
