LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Tomas Kouba said:
> > Cores=2, Benchmark=9.905-HEP-SPEC06
>
> Is the space after the comma mandatory? The lcg variables description
> at
> https://twiki.cern.ch/twiki/bin/view/LCG/Site-info_configurati
> on_variables
> differs from the document at
> http://twiki.cern.ch/twiki/pub/LCG/WLCGCommonComputingReadines
> sChallenges/WLCG_GlueSchemaUsage-1.8.pdf
Flavia's document (the second link) is the definitive source. In this
case I would expect the space not to be significant but it's probably
safer to include it since it's in the document - I cc'ed Flavia to see
if she has any comment.
> We have quite heterogeneous site, what is then meant by
> "typical WN"? Our Cores can
> be 1,2,4 or is it expected to publish a (rounded) average?
If your nodes are very different then it may be a good idea to publish
them as more than one subcluster, although it's quite a lot more work
(you then need separate queues/CEs as well). The subcluster model was
supposed to be that all nodes in it are the same, so if they aren't the
best you can do is publish either a typical or minimum value. If most of
your nodes have e.g. 2 cores then publish 2 - I don't think even a
rounded average, e.g. 3, is a good idea since you probably don't have
any 3-core nodes!
> A similar question applies for Benchmark. The document uses
> the term "CPU power",
> so is the Benchmark meant for whole CPU or just per core?
The unit is per core - the documentation on the hepspec benchmark should
explain that. If you're still using specint that was defined before we
had multi-core CPUs but you should try to estimate a per-core value. The
basic point is that most grid jobs, certainly for HEP, are run as one
job per core so that's the measure of power which is important.
Stephen
--
Scanned by iCritical.
|