[torquedev] Torque Semantics
knielson at adaptivecomputing.com
Fri Jun 11 10:03:50 MDT 2010
On 06/11/2010 03:30 AM, "Mgr. Šimon Tóth" wrote:
> I think, that we have a problem with poorly defined semantics. There
> should be a separate document describing what exactly does Torque
> consider, what parts of Torque are affected and how.
> For example requests like -l mem=1G are ignored during the scheduling
> process, but enforced on the node.
> We should ignore any behaviour that is produced by external schedulers
> that are designed to run in 1:1 (server:scheduler) configuration,
> because such schedulers override Torque semantics by sending very
> specific job run requests.
Yes. It seems over time keywords like procs and ncpus have been added to
help schedulers better utilize the cluster but TORQUE has not been
modified to keep up. We need to examine the other resources such as mem
to make sure they are uniformly interpreted as well.
More information about the torquedev