Ian Tomalin (21/04/06)

Updates to the data rates quoted in CMS NOTE 2002/047
& new proposal for FED-FRL merging

Known corrections to data rates:

1) In Table 2a of CMS Note 2002/047, the hit occupancy was substantially higher in end-cap disks TEC7-9 than in disks TEC1-3. This is now known to be due to a bug in the Monte Carlo production, which resulted in low momentum (1 MeV) photons incorrectly producing hits in the Tracker. (The flux of such photons is highest near the ECAL endcap, where they are produced by neutron absorption).

   With the latest Monte Carlo, this effect disappears. It is therefore believed that the occupancy in disks TEC4-9 will be similar to that in disks TEC1-3.

2) In Section 3 of CMS Note 2002/047, it was assumed that the FED event header (defined by the contents of the event when no hits are present) is 112 Bytes. A more recent estimate is 170 Bytes, with the increase being due to: a) One extra bit for each of the 192 APVs read by a FED, to indicate if it is in error; and b) Padding with empty bits at the end of the data sent by each of the 8 front-end FPGA chips, so that its data length is a multiple of 64 bits and so can easily be transmitted via the S-Link to the DAQ. At 100 KHz L1 trigger rate, this increases the data rate by 6 MB/s/FED.

3) It is suspected that the time resolution of the front-end electronics is somewhat worse than the 12.1 ns assumed in the note, though a precise test-beam measurement has not yet been made. This may increase the occupancy everywhere by ~14 %.

Corrections to FED-FRL mappings:

The DAQ group request that the number of FRLs assigned to the Strip Tracker be reduced from 272 to 256. Their FRLs can merge data from only 2 FEDs (decision May 2006). I therefore propose to revise the FED-FRL original mapping given in Table 4. Considerations when doing so are: a) Not to merge FEDs from different trigger partitions; b) To balance the data rates into FRLs as much as possible during high luminosity p-p collisions; c) To note that in low luminosity p-p collisions, the contribution of the FED headers to the data rate is substantial, which renders the data rate less sensitive to the hit occupancy. Furthermore that in Pb-Pb collisions, if they are taken in FED raw data mode, the data rate is insensitive to the hit occupancy. To balance the data rates in the FRLs in these cases, it is therefore wise not to merge to many FEDs into one FRL; d) It is desirable to minimize the S-Link cable length between the FEDs and FRLs, taking into account the known FED layout in the racks .

Using the notation of Table 4 of CMS Note 2002/047, it is proposed to change the merging schemes of the TEC layers, leaving the TIB, TID and TOB unchanged.  The revised estimates of the hit occupancy in the TEC allow for a reassessment of the merging there. The proposed FED-FRL new mapping for the TEC is:

For each of the two end-caps

Merge FEDs from TEC(1-3) + TEC(1-3);        16 FRLs in this category;      High Lumi p-p data rate/FRL = 144 MB/s
                             TEC(1-3) + TEC(4-6);          8                                 ;                                                      144 MB/s
                             TEC(4-6) + TEC(7-8);        16                                 ;                                                      144 MB/s
                             TEC(4-6) + TEC9      ;          8                                 ;                                                      144 MB/s 

The data rates quoted here include corrections (1) and (2) quoted at the top of this page, but not (3) as it remains uncertain. With these changes, the number of FRLs required is reduced from 272 to 240.