OK, since we have short runs confirming the times, we probably can write the
report without another long run, still....

It would be good to have a long run with the logs archived, so we can go
back to them later if necessary.  So, I suggest we go ahead and fix this and
do one LAST long run for DC1 purposes...

> From: jmyers <[EMAIL PROTECTED]>
> Reply-To: LSST Data Management <[email protected]>
> Date: Fri, 27 Oct 2006 13:53:15 -0700
> To: LSST Data Management <[email protected]>
> Subject: Re: [LSST-data] DC1 status [DataChallenge]
> 
> Ray and all others,
> 
> I just double-checked with Tim and our nodelist, etc.  We did run on 16
> nodes with 192 CCDs, which means 192/32 = 6 cores / CCD, which is twice
> what we meant to do.  This would explain why the runtimes looked longer
> than I remembered.  It actually means that (since we have perfect
> parallelism at this level, really) we can expect 65-70 sec per image.
> 
> -Jon
> 
> Ray Plante wrote:
>> On Fri, 27 Oct 2006, jmyers wrote:
>>> As it stands, I'm a little bit confused right now - Tim says we
>>> should have 3 stageslices/CPU core.  With 32 hosts x 2 cores x 3
>>> slices, that should be 192 stageslices.  This much looks good:
>> 
>> Actually is 16 hosts x 2 processors, so yes 6 slices per processor.
>> 
>> (Hmm, I guess if you were thinking 3 slices per processor, then we
>> should have done even better.)
>> 
>> cheers,
>> Ray
>> _______________________________________________
>> LSST-data mailing list
>> [email protected]
>> http://www.lsstmail.org/mailman/listinfo/lsst-data
> 
> _______________________________________________
> LSST-data mailing list
> [email protected]
> http://www.lsstmail.org/mailman/listinfo/lsst-data
> 

_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data

Reply via email to