Is anybody using CPU scaling with SGE?
thanks
Duncan
Duncan Rand wrote:
> Hi Yves
>
> OK, that seems simple enough; hopefully we can do something similar sith
> SGE.
>
> thanks
> Duncan
>
> Yves Coppens wrote:
>> Hi Duncan,
>>
>> I added the following lines (on all workers):
>>
>> $cputmult 0.633
>> $wallmult 0.633
>>
>> to the pbs config file in /var/spool/pbs/mom_priv.
>>
>> The 0.633 factor is the ratio of spec of the worker nodes with the
>> spec I publish.
>>
>> Yves
>>
>>
>> On Tue, 15 Jul 2008, Duncan Rand wrote:
>>
>>> I should have CC'd tb-support instead of dteam - if anyone replies
>>> please CC tb-support.
>>>
>>> thanks
>>> Duncan
>>>
>>> Duncan Rand wrote:
>>>> Hi Yves
>>>>
>>>> How do I implement 'CPU scaling' - at QMUL we have two sub-groups of
>>>> x86_64 nodes behind one CE with differing KSI2k values and I want to
>>>> distinguish between the two.
>>>>
>>>> thanks
>>>> Duncan
>>>>
>>>> Addition of extra clusters to an existing LCG site
>>>>
>>>> There are three ways to add an extra cluster to an existing LCG
>>>> site, namely:
>>>>
>>>> 1. Using CPU scaling
>>>> * easy to implement, but
>>>> * only the CPUs characteristics of the clusters can differ
>>>> * the same VOs must be supported on all clusters
>>>> * the site information provider does not reflect the true
>>>> physical caracteristics of the clusters
>>>>
>>>> https://www.gridpp.ac.uk/wiki/Birmingham
>>>
|