On Thu, Nov 19, 2009 at 11:47 AM, Douglas McNab
<[log in to unmask]> wrote:
> Hi Jeremy,
>
> I was just reading the agenda and I see a section on gstat2 and checking
> (availability report) cpu figures.
> I was going through this with Steve Traylen and going through the figures
> and Nagios alarms for
> http://gstat-prod.cern.ch/gstat/site/UKI-SCOTGRID-GLASGOW/
>
> Publishing is/should be straight forward. Physical is sockets/CPU's and
> Logical is Cores.
>
> If you have multiple core worker nodes these values should not be the same.
> As you suggest in the agenda some sites are publishing incorrectly or have
> the same values twice. If you have more than one CE for the same cluster
> the easiest thing to do is not publish from them all. You can override the
> values from all CE's bar 1 to be zero or else they will all be added
> together. This can be done in as a node specific override in YAIM. i.e. At
> Glasgow we have 4 CE's. 3 of which publish 0's and 1 which publishes
> correctly.
>
> In order to satisfy the CE nagios tests from the new gstat page you site
> must satisfy the requirement Physical * Cores = Logical If you have two
> different classes of machine these really need to be published separately.
Or you can publish a decimal for Cores. Will be stopped as a possibility
once multi yaim supported.
> At Glasgow have two classes of machines with different cores/SPEC values and
> to satisfy the requirement above we have had to handcraft an additional LDIF
> section for an other sub-cluster and publish both classes of machine. This
> appears to be the only way to pass the central nagios ce tests. (our
> outstanding ce errors relate to the cream tests which are in fact wrong,
> there are also some issues with the SE tests)
>
> example publication:
>
> subcluster1
> GlueSubClusterPhysicalCPUs: 280
> GlueSubClusterLogicalCPUs: 560
> GlueHostProcessorOtherDescription: Cores=2,Benchmark=7.68-HEP-SPEC06
>
> subcluster2
> GlueSubClusterPhysicalCPUs: 338
> GlueSubClusterLogicalCPUs: 1352
> GlueHostProcessorOtherDescription: Cores=4,Benchmark=8.15-HEP-SPEC06
>
> This appears on the new GStat as Physical: 618 and Logical: 1912
>
> If you want sub clustering the only way to do the required work is by
> setting the variables in YAIM on the CE and then manipulating the produced
> LDIF's by hand.
> Steve T said that working sub-clustering in YAIM has been written but not
> released. It is currently being tested again since it's been a while since
> it was written.
>
> Cheers,
>
> Dug
>
>
> 2009/11/19 J Coles <[log in to unmask]>
>>
>> Dear All
>>
>> The is a reminder of the meeting scheduled for today at 11:00:
>> http://indico.cern.ch/conferenceDisplay.py?confId=73560. The EVO password is
>> as usual dteam.
>>
>> There are several important areas to cover and some rely on the status
>> updates put in the wiki pages:
>> http://www.gridpp.ac.uk/wiki/Site_status_and_plans. If your site entry is
>> incomplete or out-of-date please could you update it before the meeting?
>>
>> After today's standard monthly meeting we are moving to a more open
>> deployment team meeting (11am on Tuesdays) which is open to everyone every
>> other week (or as required). The reason for the change is to provide a more
>> frequent forum for resolving problems during initial data taking. If you
>> have questions then we can discuss them during today's meeting.
>>
>> Kind regards,
>> Jeremy
>
>
>
> --
> ScotGrid, Room 481, Kelvin Building, University of Glasgow
> tel: +44(0)141 330 6439
>
--
Steve Traylen
|