 |
 |
Don't be so pedantic Ewan. By 'only record raw cpu time' I meant that they do not record any other type of cpu time such as normalised, not that they don't record any other information at all. Number of jobs would be fine if job lengths were random across sites but there is a strong correlation between the physics datasets stored at a site and the time needed to analyse them. Hence, ATLAS (not me) expressed a desire to weight by normalised cpu time. John -----Original Message----- From: Testbed Support for GridPP member institutes [mailto:[log in to unmask]] On Behalf Of Ewan MacMahon Sent: 05 April 2011 00:28 To: [log in to unmask] Subject: Re: HEPSPEC06 numbers for GridPP metrics > -----Original Message----- > From: Testbed Support for GridPP member institutes [mailto:TB- > > Alistair, that is the plan. The problem is that panda and the ATLAS > dashboards only record raw cpu time. AIUI that's not quite true - they record jobs done, and how much (raw) CPU time they took. Clearly, if you use raw CPU time then that would favour sites with slow CPUs, but if you ignore the CPU time and just do it by total jobs done then that problem disappears. Others may appear to replace it, of course, but it's certainly not clear to me that there are any particularly serious ones. Ewan