On 09/03/2012 17:16, Alessandra Forti wrote:
>> If there was an easy way to measure actual job throughput with varying numbers
>> of hyperthreads in use, that could be an interesting comparison.
>>
> Our job throuput definitely increased last May when we enabled HT. You can look
> at the historical dashboard.
>
Oh yes, I'm sure that would be the case. What I was wondering is to what extent
the bulk of the increased throughput actually comes from the use of the first
few hyperthreads, and if there's a point, in real usage, where increased use of
hyperthreads typically becomes detrimental to throughput (relative to use of
fewer hyperthreads).
--
Robert Fay [log in to unmask]
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/
|