Last week PIC asked:
In the GridMap monitoring (http://gridmap.cern.ch) if one clicks the "show SI2k" button in the "topology view" section, the sites are scaled wrt the "total cpus" value in a SI2k units, which looks as computed just multiplying the number of job slots published times the GlueHostBenchmarkSI00. As most
of the clusters are not homogeneous, this is not correct.
GlueHostBenchmarkSI00 is just the value to which internal accounting
is normalized.
Answer: If you scale CPU speed to some value then you should also scale your SubCluster's Total and Logical CPU count so that your total power is reflected.