[log in to unmask]"
type="cite">
LHC Computer Grid - Rollout [mailto:[log in to unmask]] On
Behalf Of Alessandro Paolini said:
thanks Stephen, I've read the bug, but in my opinion
GlueCEPolicyAssignedJobSlots should be different from GlueCETotalCPU
since this second parameter doesn't reflect any more the real capacity
of make run jobs of a certain site (the relation 1job per cpu is not
valid any more since now the capacity of a WN is measured in terms of
cores).
The history of this is that TotalCPUs (and FreeCPUs) was just a bad
choice of name, they always counted job slots but it was assumed at some
point that we would have 1 job per CPU. For glue 1.2 we decided to
rename the attributes to make the meaning clearer, so AssignedJobSlots
== TotalCPUs and FreeJobSlots == FreeCPUs by definition (they all count
slots). If it had worked properly we would have removed the *CPUs
attributes several years ago. However, in practice we never managed to
do it - with hindsight it would have been better to not even try, and
just live with the bad names!
You're right that there are some inconsistencies with multicore nodes.
This is partly tackled with the new usage document:
https://twiki.cern.ch/twiki/pub/LCG/WLCGCommonComputingReadinessChalleng
es/WLCG_GlueSchemaUsage-1.8.pdf
However, in glue 1.3 it's impossible to get it completely right, we just
don't have the right structure. For glue 2 we tried to improve the
situation, there's more flexibility to describe the configuration.
since I've seen that many lfs sites have set
GlueCEPolicyAssignedJobSlots (I don't know how), I thought that it would
have been possible for torque too, but maybe we have still to wait .-)
LSF has its own info provider. However, it was wrong there too!
Stephen
Hi,