[Lustre-discuss] Interpreting stats files

[Brock Palen]

On Nov 7, 2014, at 6:26 PM, Dilger, Andreas <andreas.dilger at intel.com> wrote:
> 
> On 2014/11/07, 4:06 PM, "Dragseth Roy Einar" <roy.dragseth at uit.no> wrote:
> 
>> Many thanks for the quick replies.  lltop seems to be a good start for a
>> tool to single out the heaviest IO users. Just need to create a wrapper
>> that maps the node names to torque jobids.
>> 
>> Have a nice weekend!
> 
> If you have Lustre 2.4 or later, you can enable the "Jobstats" aka "JobID"
> functionality in Lustre and it will handle the mapping of RPC statistics
> to Torque jobids already.
> 
> This is described in the Lustre User Manual.
This is cool never seen it before!  

https://build.hpdd.intel.com/job/lustre-manual/lastSuccessfulBuild/artifact/lustre_manual.xhtml#dbdoclet.jobstats

Question though, is it really per job? Or is it per node combining multi node jobs into one set of stats?

In our case we allow multiple jobs on a node, would job A and job B on the same node each have their own stats? Or will their stats overlap?

> 
> Cheers, Andreas
> -- 
> Andreas Dilger
> 
> Lustre Software Architect
> Intel High Performance Data Division
> 
> 
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss