[Vagnekman-users] Memory Allocation Defaults are now in place on Vagn

[Johan Raber]

Dear Vagn Users,

In anticipation of the coming expansion of the cluster with two new
nodes with 32 Cores and 256 GB memory, default memory limits per CPU
core have been introduced on Vagn.

The default has been set to 4000MB per core allocated, roughly 1/8 of
the least memory equipped machines on Vagn, thereby fitting 8 one-core
jobs on any current machine in the cluster.

For instance a two-processor job will by default get 8000 MB of memory
allocated. To allocate more (or less) memory you will need to specify
this on the command line or as an SBATCH directive in the job submission
script. The switch is "--mem-per-cpu" and would for instance be invoked
with interactive as follows:
$ interactive --mem-per-cpu=8000 #Requesting 8000MB of memory per CPU 
core used

Consequently, any jobs submitted in the "noshare" partition will get
32000 MB memory allocated by default, so if you want to make use of the
current and coming "fat" nodes you will need to include the
"--mem-per-cpu" switch set appropriately high. Currently on a6 (the only
fat node at present) you would need to set "--mem-per-cpu=8000" while
submitting to the "noshare" partition in order to fully utilize a6's
memory.

These memory limits are enforced, which means that if your application
consumes more memory than requested, it will be killed by the scheduler
(SLURM) automatically. This is introduced mainly as a protection for all
of you who use the "share" partition and will hopefully rid us of most
memory exhaustion crashes going forward at the expense of some killed
jobs which violate their memory allocation.

For more information please consult the manual of either the sbatch or
the srun commands or if this doesn't suffice, mail support. The web user
guide will be updated shortly.

Best Regards
Johan Raber

-- 
Johan Raber
E-science coordinator
National Supercomputer Center, Linköping University
SE-581 83 LINKÖPING, SWEDEN

tel. +46 (0) 13 282619
mob. +46 (0)73-6569371