hpc:hpc_tubit:slurm_usage:memory_limit

Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen angezeigt.

Link zu dieser Vergleichsansicht

Beide Seiten der vorigen Revision Vorhergehende Überarbeitung
Nächste Überarbeitung		 Vorhergehende Überarbeitung
hpc:hpc_tubit:slurm_usage:memory_limit [2019/06/21 16:30] – kraushpc:hpc_tubit:slurm_usage:memory_limit [2019/06/24 13:07] (aktuell) – kraus
Zeile 1: Zeile 1:
 +====== Memory Limits ======
  
 +
 +===== Type of Memory Limits =====
 +
 +  - Resident memory (RSS): amount real memory in use by process
 +  - Virtual memory (VMem): amount of virtual memory pages allocated by process
 +  - In the current cluster setup, only the limit for the real memory has to be taken into consideration for jobs (by parameter <html>--mem</html>).
 +\\
 +===== Properly setting Memory Limits =====
 +
 +  - The demands of your jobs for real memory (RSS) have to be carefully probed (e. g. doing test runs or or by experience ;-)) and defined.
 +  - If possible, please declare memory limits within the main input (control) file of your job (e. g. <html> <span style="font-family:courier;font-size:130%">Gaussian</span> knows about the parameter <span style="font-family:courier;font-size:130%">%mem=</span></span></html>).
 +  - SLURM defines a proportional factor between RSS and Vmem: ''VSizeFactor*RSS = VMem''.
 +  - VMem limit (VSizeFactor=0) has not to be taken into consideration.
 +  - <del>RSS limit: ''RSS = VMem limit''/''$(scontrol show config|grep VSizeFactor|awk '{print $3;}')''  and then passing via switch ''--mem='' to sbatch.</del>
 +
 +
 +\\
 +<WRAP rightalign>
 +date of revision: 06-24-2019 © kraus
 +</WRAP>