Morecombe?
The number or configuration of cores shouldn't affect APEL at all. You should be publishing the power of one core and (if appropriate) the scaling of your cpu numbers by the batch system.
The number of cores etc might affect your gstat publishing of total capacity but not APEL or ATLAS.
The things to check are: SI2K (calculated from HS06); CPUScaling Reference (if your batch system changes anything); and what your batch system actually does for normalisation.
John
-----Original Message-----
From: Testbed Support for GridPP member institutes [mailto:[log in to unmask]] On Behalf Of Matt Doidge
Sent: 19 April 2011 08:39
To: [log in to unmask]
Subject: Re: Operations team (& sites again) meeting @ 11am in EVO
I'm afraid I have to send my apologies also, I'm going into the West today.
Pete Love will hopefully be able to attend from Lancaster, but to give
you the short summary it looks like APEL is seeing half as many CPU
hours accounted for atlas as atlas themselves are reporting used.
However they're both seeing a similar amount of jobs done, so it looks
like the APEL hours are being scaled down by a factor of 2, We're not
sure what's causing this (the only bad things about our publishing is
that our physical CPU/logical CPU numbers are a bit screwy - for example
our cluster of 64 Dual-quad core boxes is presented as 64 x 8-cores
rather then 128 x 4-cores). Could this be a problem? The last time I
tried to fix it our publishing went all funny so I switched back.
To add to our fun on the 6th of April one of our clusters simply stopped
pushing out it's accounting records, it looks like the log parser for
lsf isn't behaving on that node.
Both of these issues could do with someone more knowledgeable in the
ways of APEL to take a look, but due to the fact that I'm about to go on
leave for a week I've held off on opening up support tickets until I get
back.
Have a good Easter all,
Matt
Ewan MacMahon wrote:
>> -----Original Message-----
>> From: Testbed Support for GridPP member institutes [mailto:TB-
>> [log in to unmask]] On Behalf Of J Coles
>>
>> This week is the meeting slot usually set aside for the more operations
>> team focussed discussions; but this week we want to spend half the meeting
>> discussing the GridPP metrics that will be used for the Tier-2 accounting,
>> so all site admins are encouraged to attend.
>
> I'm not going to make it; I have to be somewhere else. If we're going
> to have meetings off the usual schedule to discuss important stuff it
> would be good to have more than thirteen and a half hours notice.
>
> Ewan
|