body p { margin-bottom: 0cm; margin-top: 0pt; }
I had issues with AuthType=auth/none in the past, although not in
single node systems. I would advise using munge even for single
node, as that is the (much) more widely tested env.
Two possible issue -
1. FastSchedule=1
You didn't provide RealMemory amount, but with fastschedule, slurm
doesn't probe for amount of memory, which can lead to overallocation
in memory-bound jobs.
2. SelectType=select/linear
with a single node - this would allocate the entire node always.
I would first set the parameters to
FastSchedule=0
SelectType=select/cons_res
SelectTypeParameters=CR_CPU
NodeName=theseus CPUs=1 RealMemory=1 State=UNKNOWN
and check again (the "1" values for cpus and memory will prevent
setting the node state to down in case the slurm daemon counts
differently from you).
Often, you can see the reason in the log files. Try setting the
debug level to higher values, and if possible upload the error from
the log to the list.
SlurmctldLogFile=/tmp/slurmctld.log
SlurmdLogFile=/tmp/slurmd.log
Hope this helps
On 13/07//2012 22:51, Jones, Reese
wrote:
hi
this is very much a newbie question:
* is there a (simple) example configuration file for a single node SMP?
* does a single node installation require MUNGE?
thanks,
Reese
here is my unsuccessful attempt at creating the conf file
