On 10/17/2012 03:29 PM, Alastair Dewhurst wrote:
> I know that at least one site, normalises wall time depending on the speed of the machine.
Hi Alastair,
We also normalise (scale) wall and cpu times at Liverpool.
As Stephen says, there are many sites that do this. Heterogeneous
sites (the majority AFAIK) that do not have a CLUSTER node (most don't
AFAIK) must scale time to a benchmark reference value to ensure that
publishing for both power and accounting is correct (e.g. see
https://www.gridpp.ac.uk/wiki/Publishing_tutorial ). Queue time is
subject to normalisation when jobs run in a node, i.e. the value must
multiplied by the scaling factor of the node-type to give the real-time
max job duration. Whatever scheme you devise would have to take account
of that somehow.
Note: if the near-real-time information system contains poor data, it is
right to fix the system (i.e. if the data is copied and duplicated
elsewhere, it would go stale over time and become useless).
Cheers,
Steve
--
Steve Jones [log in to unmask]
System Administrator office: 220
High Energy Physics Division tel (int): 42334
Oliver Lodge Laboratory tel (ext): +44 (0)151 794 2334
University of Liverpool http://www.liv.ac.uk/physics/hep/
|