The document the twiki should have been populated from is at https://twiki.cern.ch/twiki/pub/LCG/WLCGCommonComputingReadinessChallenges/WLCG_GlueSchemaUsage-1.8.pdf
The logic for APEL is
1. If CPUScalingReferenceSI00 is published, use that to normalise.
2. If not, use GlueHostBenchMarkSI00
3. As Steve says, if you have actually measured HS06, then (a) publish the value in GlueCECapability and (b) calculate GlueHostBenchMarkSI00 from it.
Publishing HS06 shows you actually measured it. The APEL record doesn't actually specify a unit for the benchmark and the database just contains a number so migrating internally to HS06 in APEL is difficult.
John
-----Original Message-----
> I've been having a read around and from various references, including this:
> https://twiki.cern.ch/twiki/bin/view/LCG/Site-info_configuration_variables
> I'm pretty sure that this value is:
> - Not the one used by APEL,
> - Not altered at sites using CPU time scaling,
> - Is either accurate, or a weighted average across the nodes behind a given queue,
> - Is the value that ATLAS are asking for.
>
> Now, if I'm right then Andrew can just pull this from the information system
> and the rest of us can stop thinking about it for now. So, anyone think I'm
> not right?
Is there a clearer definition of the "Benchmark" variable
than in the above Twiki URL?
"The second value of this attribute MUST be published only in the case
the CPU power of the SubCluster is computed using the Benchmark
HEP-SPEC06."
ie cores vs logical CPUs? ie including hyperthreads or not?
It would also need cross-referencing from the CE-queue names in the
Dashboard to the CE host+domain names in the BDII, but that should be
straightforward since (as far as I can see) the CE host names all match
in both places.
Cheers,
Andrew
--------------------------------------------------------------
Dr Andrew McNab, High Energy Physics, University of Manchester
|