D0sim bg data rate estimate for the UTA Monte Carlo farm
David Adams
26sep00 1430

The pileup package includes background for the preceeding 3 (or is this 4?)
crossings in addition to that in the crossing of interest.  So if we use an
average of 2.5/xing, we need 10 bg/event.

In one run of d0sim from psim01.1 we processed 200 events in 149 minutes
on a 500 MHz processor.  If we scale with clock speed from 500 to 866 MHz,
we get 2.32 events/minute.  This give a rate per CPU of

R = 2.32 events/min * (10 bg/event) * (0.54 MB/bg)
  = (0.39 bg/sec) * (0.54 MB/bg)
  = 0.21 MB/s

Counting the old machines, UTA will have a total of 40 CPU's (866 MHz
equivalent). The total rate is

Rtot = 40 * (0.39 bg/sec) * (0.54 MB/bg)
     = (15.6 bg/sec) * (0.54 MB/bg)
     = 8.4 MB/s

I assume there is no problem accessing the disks at this event rate, i.e
the latency is tolerable. The throughput of of 8.4 MB/s might be
acheivable.

If d0gstar takes twice as long as d0sim, then our normal production mode
would have d0gstar running on 2/3 of the machines and d0sim on 1/3. Then
the bg data rate drops to 2.8 MB/s.  This seems reasonable.