CPUScalingReferenceSI00 is optional. If you don't scale your CPU you don't need to publish and APEL will use the SI2K/core published value. I don't think there is any harm in including it if the values are the same.
John
> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Stephen Burke
> Sent: 07 August 2012 16:15
> To: [log in to unmask]
> Subject: Re: Finalising Cream Settings
>
> Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of emyr.james said:
> > We will initially have 8 identical nodes...
>
> So presumably there is no scaling in the batch system, i.e. all time
> limits are in real time? Also, are you running one job per core and no
> hyperthreading?
>
> > CE_PHYSCPU=16
> > CE_LOGCPU=96
> > CE_SMPSIZE=12
>
> Yes - 16 chips, 96 cores total, and 12 cores per WN.
>
> > CE_SI00=3438
> > CE_SF00=3438
>
> SF00 is unused and undefined, but it's probably reasonable to just
> duplicate SI00.
>
> > CE_CAPABILITY="CPUScalingReferenceSI00=3438"
>
> Yes, if there's no scaling in the batch system. Note that if you
> install glexec you should also publish that in the CE Capability, and
> at least for LHC VOs you should indicate the share you give to each VO
> with something like "Share=atlas:50".
>
> Stephen
>
> --
> Scanned by iCritical.
--
Scanned by iCritical.
|