Make New Entry, Make Followup Entry
User name Bob/Alex
Log entry time 10:13:02 on March 4,2011
Entry number 350036
This entry is a followup to: 350012
Followups:
keyword=deadtime checks, part IIIHere are the readout deadtimes from the crates used
in UB2ArmNTS11 and B2ArmNTS11 configs
ROC1 ... 78 usec
ROC2 ... 65 usec
ROC3 ... 65 usec
ROC4 ... 95 usec
ROC10 ... 200 usec !! <<-- bottleneck
BBTS1 ... normally 2 usec, but once per 100 events its 350 usec
(so the average should be around 3.5 usec from BBTS1)
(unbuffered) Trigger Supervisor Busy .... 220 usec DT. It is dominated by ROC10.
We don't know much about BBTS11, but it seems to read scalers
once per 100 events. See fig 1 for an event dump of ROC29 "long" event.
(ROC29 is synonymous with BBTS11).
We also found that the buffered mode was NOT buffered.
The string should be "buf". Admittedly this would be hard
to know, but on the other hand it could have been discovered
by looking at daLogMsg from the crates. So, we were never buffered.
Now look at deadtime for the 2 modes. Note, we are running with
loose triggers at 2 kHz (observed rate) and 3.6 kHz (input).
Note (1 - (2.0/3.6)) = 0.44. (44% DT)
And datamon says 40 %. This is Unbuffered (UB2ArmNTS11).
Now run buffered mode. B2ArmNTS11. (I did the DB2Cool translation
after changing this config to actually do buffering).
Wow ! Buffering helps !
Trigger supervisor DT drops to 21 usec
(buffered) Datamon reports DT = 7 % !!!! (it was 40 % unbuffered)
We're going to check ROC10 because it seems to be the slowest
at the moment.
FIGURE 1
