> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of John Gordon
>
> But surely all analysis jobs are not equal? Production simulation could be
> judged on number of events generated but analysis will vary greatly
> depending on the physics dataset and the size of the sample. Surely there
> needs to be some weighting?
>
Quite possibly, but it might be safe to assume that job to job
variations would average out - unless there's a consistent reason
that one site gets little jobs while another gets big ones (?).
If we could make some sort of 'jobs actually done' metric work it
would neatly encapsulate all the properties of the site, give
credit for things like fast IO storage, better networking,
hyperthreading, turbo mode, and whatever else a site can scrounge
up to make the jobs run quicker, while also avoid having to worry
about things like the details of hardware in each node.
Ewan
|