Chris,
> So if we publish twice the raw CPU hours and half the HS06, we claim
exactly the right number of Hepspec hours don't we?
QMUL discrepancy between Atlas and Apel cpu hours is among those
discrepancies that triggered all this.
> You are still implying that what we were doing is incorrect.
why would I do that? I suggested myself at the ops meeting few weeks ago
you should do what you are doing now eliminating the cpu hours factor 2
discrepancy and publishing the best HS06 value as possible rather than
half that value.
> I have changed it to scale to the value of our most recent CPUs -
mainly to avoid this sort of discussion if I'm honest. Is that "correct"
in your eyes?
it eliminates an "unexplained" discrepancy between Atlas and Apel
accounting systems. So yes, I think it is certainly a better way to
account for QMUL HS06 hours.
cheers
alessandra
>> There was a factor 2
>> that could be moved around. It seems you have corrected that now.
>
> You are still implying that what we were doing is incorrect. I dispute
> that[1] - and fail to understand why you think it is incorrect. Can
> you explain how this can make any difference to the resource usage we
> publish to APEL - or anywhere else?
>
> I have changed it to scale to the value of our most recent CPUs -
> mainly to avoid this sort of discussion if I'm honest. Is that
> "correct" in your eyes?
>
>
> Chris
>
>
> [1] What we were doing that was incorrect is publishing the available
> capacity at 1000hepspec. Thanks to Liverpool's blog on the subject
> I've corrected this
|