[Beowulf] Performance metrics & reporting

[John Leidel]

We report similar metrics, but do so in a very project oriented manner.
We have fairly strict project allocation guidelines that dictate
accurate system accounting.  Depending on which batch system you're
going to use, this can be fairly easy data to collect.  

The PBS accounting records are not terribly easy to read.  Our
accounting scripts utilize a simple state machine to pull the necessary
data.  GridEngine accounting records are very easy to grab data from.
Simple colon delimited format, one job per line.  Sun also released the
ARCo accounting tool.  [They released it open source immediately
following me completing development on just such a tool....]  It will
allow you to poll useful information out of the accounting records w/o
doing a bunch of scripting.  I believe Altair is working on releasing a
similar product for PBSPro [but not free].  

At any rate... hope this helps :-)   

--john

On Sat, 2008-04-05 at 13:20 -0500, Gerry Creager wrote:
> We're interested, as Matt said, in what folks are using.  We're ramping 
> up a new throughput cluster on our campus that'll be multi-disciplinary, 
> and we have several groups of stakeholders.  We want to report various 
> statistics, such  as number of jobs, average number of nodes, average 
> number of cores, average runtime, etc., plus distributions thereof.  And 
> anything else that's pointed out as "interesting".
> 
> gerry
> 
> John Leidel wrote:
> > what sort of data are you looking for? 
> > 
> > On Fri, 2008-04-04 at 09:41 -0600, Matt Lawrence wrote:
> >> What do folks recommend for gathering and reporting performance data on 
> >> Linux (CentOS 5) systems?
> >>
> >> Since I have done a lot of this on AIX, my initial thought is to use the 
> >> Linux version of nmon and nmon2rrd or nmon2web to generate web pages of 
> >> performance graphs.  Since I already know the tools, wrapping it up in 
> >> scripts to run automatically would be really easy.
> >>
> >> My goal is to make it really easy to provide useful information to users 
> >> without them having to constantly ask me.
> >>
> >> Thoughts?
> >>
> >> -- Matt
> >> It's not what I know that counts.
> >> It's what I can remember in time to use.
> >> _______________________________________________
> >> Beowulf mailing list, Beowulf at beowulf.org
> >> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> > 
> > _______________________________________________
> > Beowulf mailing list, Beowulf at beowulf.org
> > To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>