Hi Vega,
I think Stephen meant something simpler: what is the difference in total
elapsed time between the following two commands:
globus-job-run my-ce.coolsite.nl:2119/jobmanager-lcgpbs /bin/coolcmd
vs
globus-job-run my-ce.coolsite.nl:2119/jobmanager-pbs /bin/coolcmd
so the injection of the job from outside is exactly the same (RB, or
g.j.r., or whatever) ... but the INTERNAL mechanics, inside the site,
are changed. So the RB is irrelevant. You are right, submitting via RB
is slower, but that's a different question :)
JT
Vega Forneris wrote:
>
> Hi,
>
> in my experience, I already had the same results:
>
> pbs-RB (LCG approach) about 2-3 min for
> simpliest job
> lcgpbs-WMS (gLite approach) about 6 minutes f.s.j.
> and
> pbs/lcgpbs (GT4 approach) about 20 seconds f.s.j.
>
> In GT4 environment you don't have any RB/WMS so the connection goes
> directly saving time. In LCG/gLite environment you pass through a
> machine (RB/WMS) and another scheduler (generally condor). Then latest
> gLite versions have some security related issues which take longer time,
> from voms extension to the feature (i.e.) on the CE which waits for more
> than a minute after the pbs job is complete for being sure it is REALLY
> finished (previous version of pbs sometimes reported as DONE jobs still
> running...)
>
> Cheers
>
> Vega Forneris
>
> +-----------------------------------------------+
> ESA-ESRIN
> Unix Systems Administrator
> Via Galileo Galilei
> 00044 Frascati (Rm) - Italy
> Phone +39 06 94180581
> Mailto: [log in to unmask]
> +-----------------------------------------------+
> Vitrociset S.p.A.
> Unix System Administrator
> Via Tiburtina 1020
> 00100 Roma - Italy
> Phone +39 06 8820 4297
> Mailto: [log in to unmask]
> +-----------------------------------------------+
>
> "I do not feel obliged to believe that the same God who has endowed us
> with sense, reason, and intellect has intended us to forgot their use."
> (Galileo Galilei)
>
>
>
> *Stephen Childs <[log in to unmask]>*
> Sent by: LHC Computer Grid - Rollout <[log in to unmask]>
>
> 21/03/2007 18:27
> Please respond to
> LHC Computer Grid - Rollout <[log in to unmask]>
>
>
>
> To
> [log in to unmask]
> cc
>
> Subject
> [LCG-ROLLOUT] pbs vs lcgpbs performance
>
>
>
>
>
>
>
>
> We've just changed over to the pbs job manager from the lcgpbs jobmanager
> in order to fit in with the RB restrictions on MPI jobs and I've noticed a
> massive speedup.
>
> A simple globus-job-run completes in < 25 seconds (compared to around 2
> minutes for lcgpbs) and a job submitted through the RB completes in just
> over 2 minutes (compared to around 5-6 minutes for lcgpbs).
>
> I thought this change wouldn't make a huge difference as I thought most of
> the delay was in the RB. Has anyone else noticed this or was there
> something horribly wrong with our lcgpbs jobmanager that was causing the
> slowdown.
>
> My productivity is going to go through the roof when going round the test
> job cycle!
>
> Stephen
> --
> Dr. Stephen Childs,
> Research Fellow, EGEE Project, phone: +353-1-8961797
> Computer Architecture Group, email: Stephen.Childs @ cs.tcd.ie
> Trinity College Dublin, Ireland web: http://www.cs.tcd.ie/Stephen.Childs
>
|