[Bioclusters] Fwd: Re: [PBS-USERS] LSF vs its "Closest Competitor"

[Ron Chen]

One correction: I submitted 100000 jobs to my test
cluster, and the size of the SGE scheduler is only
around 450MB, which is 4.5KB per job (down from 9KB I
mentioned in an earlier e-mail). This is far lower
than what Platform Computing claimed about its 
"Closest Competitor".

Also, some comments from the PBS users list:

--- Harley Gorrell wrote:
>    460k jobs so far on a 120 cpu farm.  There is
> room for
> improvement, but openpbs is hard to beat for the
> price.
> 
> harley.


--- John Kochmar <kochmar@psc.edu> wrote:
========================================================
> > Performance, Scalability, Robustness
> > 
> >                      LSF 5       Closest
> Competitor
> > 
> > Clusters               100+                 1
> > 
> > CPUs                 200000+               300
> > 
> > Jobs                 500000+             ~10000+
> > (active across clusters)
> > 
> > Fairshare Utilization  ~100%               ~50%
> > 
> > Query Time         20% better than        40%
> slower
> >                       LSF 4.2              than
> LSF 5
> > 
> > Scheduler Usage        4K/job             28K/job
> >
>
========================================================
> > 
> > I would love to hear from the people here, at
> least a
> > number of things above are not true.
> 
> I'm not sure how to interpret these numbers, or who
> they define their
> "closest competitor" (PBSPro?), but we run OpenPBS
> on Lemieux (TCS-1),
> which has 3000+ cpus (we have live spares in the
> cluster, the scheduler
> adds those spare nodes when a compute node drops
> out) in 750+ compute
> nodes, and our utilization is regularly up around
> 85% (it would be
> better, but TCS-1 runs a smaller number of large
> jobs, so there are
> under-utilized nodes as a result of scheduling holes
> and machine drains
> -- when we've allowed smaller jobs w/out draining,
> we've seen
> utilization percentages in the high 90's.)  The
> current PBS job id's are
> over 123,000, and again, this would be better if out
> users ran rafts of
> small jobs instead of large MPI jobs.
> 
> As for how they came up with these numbers ... no
> clue.  You'll have to
> ask them.  I suspect it has to due with how they
> define "closest"
> competitor.
> 
> J
> 
> -- 
> John Kochmar <kochmar@psc.edu>
> Pittsburgh Supercomputing Center
>
__________________________________________________________________________
> To unsubscribe: email majordomo@OpenPBS.org with
> body "unsubscribe pbs-users"
> For message archives:
> http://www.OpenPBS.org/UserArea/pbs-users.html
>     -    -    -    -    -    -    -    -    -    -  
>  -    -    -    -
> OpenPBS and the pbs-users mailing list are sponsored
> by Altair.
>
__________________________________________________________________________


__________________________________
Do you Yahoo!?
The New Yahoo! Search - Faster. Easier. Bingo.
http://search.yahoo.com