On 17/07/17 15:14, Alessandra Forti wrote:
> I've updated Manchester to reflect the number of slots rather than HT
> (HT info is incorporated in the first column anyway). Memory is per node.
>
Looking at the fields now ....
I can see you have given "totalHs06" and "hs06PerCore" (ignoring
obsolete "Per HW Thread").
From that, I can tell that you have a "Dual 6 cores x 2 HTs, Intel
X5650 @ 2.67GHz".
You measured that hardware and got 211.62 totalHs06. Since you say you
had 17.64 hs06PerCore, I can compute that you use 12
instancesOfTheBenchmark/slots/threads/whatever.
Fine, it's all there. If you used those nodes for VAC, you might (for
example) have to repeat the calibration with 10 x
instancesOfTheBenchmark (because VAC needs more memory per job.)
So you have given enough information to get it all (in my view, it would
be simpler to give the totalHS06 and the instancesOfTheBenchmark. But it
makes no odds, really.)
Note: At Liverpool, I actually give all three, but only two are needed.
So, do we agree - totalHs06 and hs06PerCore should be fields, and the
"slots used" (instancesOfTheBenchmark) can be determined
arithmetically? Let me know _before_ I change the table.
Cheers,
Steve
--
Steve Jones [log in to unmask]
Grid System Administrator office: 220
High Energy Physics Division tel (int): 43396
Oliver Lodge Laboratory tel (ext): +44 (0)151 794 3396
University of Liverpool http://www.liv.ac.uk/physics/hep/
|