Make New Entry, Make Followup Entry
User name R. Michaels
Log entry time 10:22:02 on March 2,2011
Entry number 349613
Followups:
- 349626 03/02/11 11:42 Hanretty FPP HV briefly turned off to see effect on deadtime
- 349635 03/02/11 12:04 R. Michaels adaql2: system check
I was asked to worry about the DAQ deadtime for the present
config which is B2ArmNTS11.
Problem: no pedestal suppression on ROC1 and ROC2.
Thus, the event sizes are about 1 kbyte and 750 bytes
respectively.
A more normal event size is like for roc 3, which is 175 bytes.
Roc 4 has 500 bytes, perhaps due to FPP. This would be the
next bottleneck after fixing ped suppression on ROC1, 2.
Chen Or is fixing the ped suppression files ~/ped/ped1.dat and
ped2.dat. He's doing this "by hand" since apparently the pedrun
config does not work. (The reason pedrun config does not work
is that there are no ADC gates for T8, which was something I
thought got fixed during the downtime prior to the experiment,
but apparently not.)
Some comments about the DAQ deadtime. It is NOT mysterious !
Our simple theory for how it works is shown in fig 1. There is
a maximum rate of the system which is determined by the Readout
deadtime of the slowest crate. Note, the frontend deadtime (also
called frontend busy) affects the slope of the approach to R_max.
This slope is small if in buffered mode, and bigger in unbuffered
mode, but one reaches the R_max either way. Also, R_max can be
affected by network performance, which is why blaster tests are
important. I recall the blaster tests were ok (10 Mbyte/sec).
One may measure the readout deadtime of each crate as follows.
Write a "1" to the output register of the SFI at the start of
the trigger routine, and a "0" at the end. This should show
a typical readout deadtime of 150 usec on a scope. As I said
above, I suspect ROC4 will be a bit higher.
This bit is written in the CRL as
[a-onl@adaql2 ~/crl]$ grep writeVme sfi4.crl
*sfi.writeVmeOutSignalReg = 0x100;
*sfi.writeVmeOutSignalReg = 0x1000000;
It was correct for ROC3 and ROC4 but not correct for
ROC1 (missing the lines) or ROC2 (slightly wrong place).
So, I have corrected sfi1.crl and sfi2.crl.
We download again and check the DAQ. Affects run 2173 onward.
FIGURE 1
