On 04/04/2011 16:03, Stephen Burke wrote:
> Testbed Support for GridPP member institutes [mailto:TB-
>> [log in to unmask]] On Behalf Of Alessandra Forti said:
>> The problem of Hyperthreading remains. At the moment Manchester and
>> QMUL
>> are definitely loosing out when the cluster is half full which is
>> often.
> If the cluster isn't full I'm not entirely sure what's being measured anyway -
well there are actually measures on the hepix web site and we got the
same results for similar architecture.
If only 12 jobs run the same HS06 value per node is measured as if HT
was off. The value per thread used (job?) is therefore much higher.
https://hepix.caspur.it/benchmarks/doku.php?id=bench:results_sl5_x86_64_gcc_412
Manchester has the same architecture as INFN PD. So for us it varies
between 8.8 and 13.7.
cheers
alessandra
|