Re. action 108 on me. I spoke to Tony, and his law of 5:1, which he
presented in Birmingham still holds, _on_average_.
So for every 5kSI2k on your farm, you should aim for 1TB of available
storage. For reference a modern opteron dual core, dual cpu has a
rating of a little _more_ than 5kSI2K.
Tony's numbers are based on the 2008 experiment requirements. In
detail, for each LHC VO, the numbers are:
ALICE 3:1 CPU:disk (kSI2k:TB)
CMS 4:1 CPU:disk (kSI2k:TB)
ATLAS 2:1 CPU:disk (kSI2k:TB)
LHCb 333:1 CPU:disk (kSI2k:TB)
Remember, for shared farms, you'll have to scale storage by the
shares for each VO.
e.g. Glasgow are buying 750kSK2K. Our nominal share for HEP is 20%,
however, we want to process lots of ATLAS jobs so we need:
750 * 0.2 (HEP Share) / 2 (Atlas disk/cpu ratio) = 75TB
Cheers
Graeme
[1] http://www.gridpp.ac.uk/gridpp14/gridpp14_overview.ppt
--
Dr Graeme Stewart - http://wiki.gridpp.ac.uk/wiki/User:Graeme_stewart
GridPP DM Wiki - http://wiki.gridpp.ac.uk/wiki/Data_Management
|