HP XC System Software Administration Guide Version 3.0

cpu time (seconds, -t) unlimited
max user processes (-u) 8113
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
Only soft resource limits can be manipulated. Soft and hard resource limits differ. For example, the following
two commands list the hard and soft limits for the size of the core file:
$ ulimit -Hc
unlimited
$ ulimit -Sc
0
If a user sets a soft resource limit with the ulimit bash shell command and the software resouce limit can
be propagated, that value is propagated to the jobs running on the compute nodes. However, this does not
override the compute node system configuration for resource limits.
The SLURM configuration file, slurm.conf, enables you to modify a resource's soft value so the jobs
running on the allocated compute nodes have the same resource limits:
• PropagateResourceLimits propagates only the specified resource limits.
The resource limits specified are applied to the job on startup. All other resource limits are determined
by the system values defined on the compute nodes.
• PropagateResourceLimitsExcept propagates all the resource limits except those specified.
All resource limits are propagated when neither parameter is specified.
Note
Users can use the --propagate option of the srun command to propagate these resource limits. For more
information on this option, see srun(1).
Table 12-3. lists the resource limits that can be set or excluded.
Table 12-3. Resource Limits Resource Limits
DescriptionResource Limit
Maximum address spaceRLIMIT_AS
Maximum core file sizeRLIMIT_CORE
Maximum CPU time in secondsRLIMIT_CPU
Maximum size of a process data segmentRLIMIT_DATA
Maximum file size createdRLIMIT_FSIZE
Maximum size that can be locked in memoryRLIMIT_MEMLOCK
Maximum number of open filesRLIMIT_NOFILE
Maximum number of processes available to a single userRLIMIT_NPROC
Maximum resident set sizeRLIMIT_RSS
Maximum stack sizeRLIMIT_STACK
In the following example, all the system limits except RLIMIT_NPROC, which determines the maximum
number of processes, and RLIMIT_NOFILE, which determines the maximum number of open files, are
propagated from the user's environment.
PropagateResourceLimitsExcept=RLIMIT_NPROC,RLIMIT_NOFILE
If you make any changes, be sure to run the scontrol reconfigure command to update SLURM with
these new settings.
If a user tries to propagate a resource limit with the srun --propagate command, but the compute node
has a lower hard limit than the soft limit, an error message results:
$ srun --propagate=RLIMIT_CORE
.
108 Managing SLURM