J. Gilmore, Dec. 2004
Here is a copy of what I sent to Attila previously for his RUWG questions.

I have added 4 notes to indicate the corrections for some minor
miscalculations.

---------------------------------------------


First, here is a reminder of our basic DAQ structure.  EMU has 8 S-Links
in the main CMS readout.  Our DCC concentrates up to 5 DDUs into 1 S-Link,
with a total of 36 DDUs.  Each DDU carries the data for 13 DMB Front-End
units (1 DMB per CSC, ~470 in total), and each DMB passes data from 5 CFEBs,
1 ALCT and 1 TMB.
	1 S-Link  <===  5 DDUs  <===  13 DMBs

** correction:  there are "13 DMBs per DDU" in the EMU baseline design **



> Do you know your maximum event size?

The maximum event size into one S-Link is ~595 KB, but this has
essentially _zero probability_ during normal data runs; it should
only occur during calibration runs.  In some cases we may take
extended calibration runs that have S-Link events as large as
1100 KB (with 16 SCA time samples per CFEB), and this would be
done at a very slow rate.  We will only perform calibration in
a dedicated calibration run,  _not_ during a normal data run.

But I think the "maximum size" isn't the only question.  It is better to
consider the "Average Event size."  Additionally, due to zero suppression
and the addition of event headers at various levels in the system, one must
separate the "average transmit size" from the "average stored size" (they
are not equal) to evaluate overall system capacity.  Event sizes below are
in Bytes.
		DMB		  DDU		S-Link
 --------------=====------------======----------======----------
 Maximum Size	9340 		121400		607200
 Avg Stored	 22		  460		 1920
 Avg Transmit	 30		  340		 1880

** correction for DDU:  "460" above should be "390" bytes **



> What is the buffering capacity of your chain (in the FED, in the
> DCC if you have one, in the Front-End)?

The EMU buffer capacity is shown below for the number of Maximum-Sized
and Average-Sized events that can be stored:
		DMB		  DDU		DCC/S-Link
 --------------=====-------------=====----------==========----------
 Max-Sized	 10		   8		    5
 Avg-Sized	4000		 2000		  1600

** correction for DDU:  "8" above should be "20",  "2000" should be "6400" **



> Can you buffer at least one ?

The EMU DAQ system has feedbacks from the higher levels to the lower
levels which allows us to maximally use our buffering capacity.
Therefore EMU is capable of storing up to 23 Maximum-Sized events,
or 7600 Average-Sized events.

** correction: "35" Maximum-Sized or "12000" Average Sized events. **



> How do you deal with large events ?

We have no problems with large events; they are handled just like
any other event.



> How long can you take data before asserting busy state through the FMM
> port? 

We don't know exactly what you mean.  We will set WARN before BUSY;
we expect that the WARN will be set very rarely.  After we set WARN
there will typically be enough buffer available to hold 10 msec of
data before we need to set BUSY.

** Note:  for EMU to get into a WARN condition in the first place  **
 *	  there must be something very a-typical happening...      *



> Can you loose data?

It is possible that a very high rate burst could cause a temporary loss
of data, but we handle this in a controlled way such that EMU does not
lose synch and all events with data loss are marked in the DAQ stream.



> Can you loose the synchronization with the TTC counters ?

Only if caused by a radiation-induced SEU, a very rare occurence.



> How do you recover/resume ?

SEU induced errors require Hard Reset.