Log entry time 07:37:54 on February 26, 2011

Entry number 348748

This entry is a followup to: 348673

There is a FAQ about deadtime which gives a list of things to check

http://hallaweb.jlab.org/equipment/daq/dtime_faq.html

Look for unusually large event sizes from some crate.  E.g. a common
problem is to leave thresholds turned off, leading to a lot of hits
from noise. I'm not at the lab, so I cannot easily check this at the 
moment.  A typical event size is 500 bytes per crate.  But this can
get much larger, depending.  Also, pedestal suppression can help.

Missing gates (leading to timeout) causes huge deadtime, but you
presumably would have noticed that.

On the SFI, I believe we write a "1" to one of the I/O ports at the
start of the readout, and then a "0" at the end.  One may put this
into a scope during cosmics running.   A normal deadtime is 100 to 150
microsec.  A bad deadtime is 1 msec.  The deadtime of the system
is quite predictable (if the network is healthy): to first order
it is DT = R * T, where R = rate (Hz) and T = the afore-mentioned
deadtime (seconds) and DT is the "deadtime as fraction".  Then the
slowest crate dominates, so if one contributes DT = 2% and another 25%,
the system DT is 25%.