LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Jeff Templon said:
> A reasonable setting which may help speed along the Glue issue is to set
> the TotalCPU counts for CEs to *zero*. This has the advantage of giving
> a definite algorithm that is implementable across all sites.
Well, in the sense that it would then be equally wrong everywhere ...
TotalCPUs is really a bad name, Sergio's proposal already suggests changing
it to TotalJobSlots which is more reflective of what it means - it should be
the maximum number of simultaneous jobs which could ever be run in that
queue. Since queues often compete for resources you would generally only get
to that maximum if the other queues were empty, and there may also be queue
policies which stop you filling the whole system, so it's just an upper
bound. You can choose to schedule more than one job per physical CPU, so
there is no particular reason to assume that 1 job slot = 1 CPU.
Stephen
|