[torqueusers] calculating cpu time
beaneg at umcs.maine.edu
Tue Oct 11 11:58:05 MDT 2005
On Tue, 11 Oct 2005, Doug Smith wrote:
> This is a basic question but I seem to be getting conflicting answers
> from people.
> When defining a queue in qmgr setting resources_max.cput = 24:00:00
> should set the maximum amount of cpu time a job can have right? Is this
> a per CPU setting or a general cpu time setting? Meaning if I submit a
> job to 16 processors should it run for 1.5 hours (24hours/16processors)
> or will it run for 24 hours on each cpu?
24 hours cumulative. Assuming all processes are under torque's control.
> I have gotten conflicting answers to this question.
> And what if any relation does resources_max.wallclock have to this
> issue? I was under the impression that wallclock time meassured the time
> from when a job was submitted to when it terminated. If you set
> resources_max.wallclock = 48:00:00 a job could be in the system for 48
> hours in what ever stat E Q R... but that 48 hours after it was
> submitted it was going to get terminated by the scheduler/resource manager.
> Is this correct?
48 hours in the R state. It could be in the Q state for 2 years -
walltime has no relation to Q time.
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers