Hi Steve,
VAC (and other cloud resources) are a problem that needs to be solved.
It is not clear to me why you are not running ATLAS jobs though. You
have an active queue perhaps that is something we should look into. It
might also account why Liverpool in October had a bigger discrepancy
than in the previous two months (you run atlas jobs) It seems also a bit
of a waste that you put your most powerful nodes in VAC if it isn't used
by all the experiments.
python get-atlas-accounting-data.py -m3 -s UKI-NORTHGRID-LIV-HEP -n
Date,ATLAS work,EGI work,(wE-wA)*100/wA,ATLAS wc,EGI wc,(wcE-wcA)*100E/wcA
2016-08,5055330,5060007,0.1%,495657,503303,1.5%
2016-09,6101453,6178377,1.3%,598225,615048,2.8%
2016-10,4466927,5479628,22.7%,437966,542616,23.9%
cheers
alessandra
On 17/11/2016 14:42, sjones wrote:
> Hi Alessandra,
>
> On 2016-11-17 12:56, Alessandra Forti wrote:
>
>> I agree in ECDF it is variable but you really should put a meaningful
>> value that averages to a meaningful HS06 number.
>
> It's complicated (i.e. impossible) to find a value that averages to a
> meaningful HS06 number because, as discussed, ATLAS Jobs do not run
> much on VAC. And VAC does not have any BDII.
>
> What that means:
>
> Case a: if I include our VAC capacity in our BDII (piggybacking the
> resources into, say, our CREAM/Torque cluster), ATLAS will perceive
> the true size and power of our site. But, because ATLAS Jobs do not
> run much on VAC, the size of our site that they perceive is more than
> the practical size that they can run ATLAS Jobs on. If our VAC nodes
> were turned off, it would not make any difference to ATLAS because
> they don't use them. And, because the average HS06 of our VAC nodes
> is different to the average HS06 of the other clusters at our site,
> and ATLAS jobs rarely run on VAC, then it follows that ATLAS jobs
> would not receive the published average HS06 and thus ATLAS work
> calculations would be further corrupted.
>
> Case b: if I omit our VAC capacity from our BDIIs (no piggybacking),
> ATLAS will perceive that our site is smaller than it actually is. But,
> because ATLAS Jobs do not run much on VAC, the size of our site that
> they perceive is practically the size of our site that runs ATLAS
> Jobs. Furthermore, as discussed, the average HS06 of our VAC nodes is
> different to the average HS06 of the other clusters at our site. Since
> ATLAS jobs do not run on VAC, it is better perhaps not to include them
> in the BDII.
>
> But whatever I do, it is "meaningless" in one way or the other. The
> system is simply not semantically expressive enough to cater for the
> needs. However, I think it's best for ATLAS to ignore the VAC clusters
> (that they don't use much) and to put the true values (no
> piggybacking) in the other clusters. Thus, ATLAS wold perceive the
> site as smaller than it actually is. But so what - the values do (more
> or less) correspond with the parts of the site that ATLAS runs jobs
> on. Thus the ATLAS work calculations should be more accurate with that
> policy.
>
> Please let me know if you want me to include (by piggybacking) the VAC
> nodes in the node counts and averages of one of our other clusters -
> I'd be happy to do it if you need me too, but I fear ATLAS work
> calculations will be further corrupted if I did that.
>
> Cheers,
>
>
> Ste
--
Respect is a rational process. \\//
Fatti non foste a viver come bruti (Dante)
|