Hi Gordon,
I completely agree with that 1-job 1-core is not going to work for long. Memory cost is dominating. We ride a slower Moore's law. We need to uitlize multiple cores per job, like one single faster CPU. I have been pushing ATLAS but there is not much action. Processing several clusters or tracks in parallel can be problematic -- they may have overlapping hits. Processing events in parallel shuld be physics safe. I have also pushed G4 (core) without success. Cheers.
Charlie
--
Charles C. Young
M.S. 43, Stanford Linear Accelerator Center
P.O. Box 20450
Stanford, CA 94309
[log in to unmask]
voice (650) 926 2669
fax (650) 926 2923
CERN GSM +41 76 487 2069
> -----Original Message-----
> From: [log in to unmask]
> [mailto:[log in to unmask]] On
> Behalf Of Gordon Watts
> Sent: Wednesday, July 25, 2007 2:15 AM
> To: atlas-sccs-planning-l
> Subject: RE: Agenda for ATLAS/SCCS Planning Meeting 25th July 2007
>
> Hi,
> As usual, I'm not at all sure that I can make this meeting
> (it overlaps with one of the physics meetings I regularly
> attend). But, I thought I'd add my two cents worth.
>
> SLAC may already have a great deal of experience at this --
> but we found that testing modern dual-head 4 core (i.e. total
> 8 core) machines took longer than normal. We had to tweak
> some Linux OS parameters in order to really get the machine
> to run properly. A caveat in this is that we are using this
> as a DAQ machine and so we use shared memory to move data
> round, and most of what we did had to do with that. But,
> none-the-less, I would allow for a longer-than-expected
> testing time if SLAC hasn't already had some experience with
> these 8 core CPUs.
>
> I don't know about you guys -- but it is seems to me that
> at the rate we are adding cores to pizza boxes our usage
> model is going to break down (i.e. one core, one job). It
> seems at the straining point right now.
>
> Cheers,
> Gordon.
>
> > -----Original Message-----
> > From: [log in to unmask] [mailto:owner-
> > [log in to unmask]] On Behalf Of
> Stephen J. Gowdy
> > Sent: Tuesday, July 24, 2007 4:50 PM
> > To: ATLAS SCCS Planning
> > Subject: Agenda for ATLAS/SCCS Planning Meeting 25th July 2007
> >
> > ATLAS SCCS Planning 25Jul2007
> > -----------------------------
> >
> > 9am, SCCS Conf Rm A, to call in +1 510 665 5437, press 1, 3935#
> >
> > Present:
> >
> > Agenda:
> >
> > 1. DQ2 Status/Web Proxy
> >
> > 2. Tier-2 Hardware
> >
> > 3. AOB
> >
> > Action Items:
> > -------------
> >
> > 070711 Stephen Find out about benchmarks for CPUs for next
> > purchase
> > 070718 Not done yet.
>
>
|