EVBSIM Status

1
EVBSIM Status

• Fermilab XDAQ test stand simulation
• Results for 1x1, 1x2, 2x1, 2x2
• Discussion
• Plans

2
XDAQ Simulation
3
EVBSIM/XDAQ Setup

• XDAQ parameters
• set Samples 10000 set triggerDisable
  1
• set EIDs 10000 set builderDisable
  0
• set eProfile 100 set FilterEnable
  0
• set BlockSize 4096 set maxFragments
  4
• set printEvery 300 set TestFragmentSize
  2048
• MaxRequests and NumGather were varied
• EVBSIM parameters
• DMA from host memory to NIC 10 bytes/132 us
• Myrinet transfer 5
  bytes/160 us
• DMA from NIC to host memory 15 bytes/132 us
• values obtained by tuning 1x1 case, starting
  from estimates

4
1x1 EVBSIM/XDAQ Results

• Ratios of EVBSIM and XDAQ bandwidths
• (in MB/s) for various settings
• R \ N 1 4 16
  64
• 1 10/101.0 25/261.0
  43/441.0 52/531.0
• 4 23/231.0 43/490.9
  55/580.9 59/611.0
• 24/251.0 43/490.9 55/580.9
• 64 24/251.0
• R number of concurrent event request packets
• N number of event requests per packet

5
1x2 EVBSIM/XDAQ Results

• Ratios of EVBSIM and XDAQ bandwidths
• (in MB/s) for various settings
• R \ N 1 4 16
  64
• 1 9.5/110.9 25/261.0
  36/301.2 37/311.2
• 4 21/221.0 32/291.1
  36/311.2 37/311.2
• 22/221.0 32/271.2 36/311.2
• 64 22/221.0
• R number of concurrent event request packets
• N number of event requests per packet

6
2x1 EVBSIM/XDAQ Results

• Ratios of EVBSIM and XDAQ bandwidths
• (in MB/s) for various settings
• R \ N 1 4 16
  64
• 1 16/190.8 36/450.9
  51/660.8 57/740.8
• 4 31/330.9 49/570.9
  58/720.8 60/780.8
• 31/330.9 49/570.9 58/720.8
• 64 31/340.9
• R number of concurrent event request packets
• N number of event requests per packet

7
2x2 EVBSIM/XDAQ Results I

• Ratios of EVBSIM and XDAQ bandwidths
• (in MB/s) for various settings
• R \ N 1 4 16
  64
• 1 17/190.9 36/430.8
  50/531.0 57/531.1
• 4 31/321.0 49/540.9
  58/591.0 60/451.3
• 31/330.9 49/550.9 58/591.0
• 64 31/330.9
• R number of concurrent event request packets
• N number of event requests per packet
• 45 XDAQ value not understood

8
2x2 EVBSIM/XDAQ Results II

• EVBSIM bandwidths now automatic like XDAQ.
• Ratios of EVBSIM and XDAQ bandwidths
• (in MB/s) for various settings
• R \ N 1 4 16
  64
• 1 17/190.9 39/430.9
  57/531.1 62/531.2
• 4 30/320.9 53/541.0
  61/591.0 62/451.4
• 30/330.9 53/551.0 61/591.0
• 64 30/330.9
• R number of concurrent event request packets
• N number of event requests per packet
• 45 XDAQ value not understood

9
Discussion

• Automatic BW results have a problem
• Sometimes BW drops when trigger rate too high
• EVM can delay sending event IDs ???
• Must be solved
• Manual BW results have a problem
• For 2x2 automatic BW is sometimes lower
• Only higher BW can be understood
• Some latency/BW setting must have been wrong
• 1x1/1x2/2x1 will be redone too

10
EVBSIM Plans

• Simple model reasonable so far
• Fix automatic BW determination
• Keep extending to 1xN, Nx1, NxN
• Compare with new XDAQ
• Try to find parameters causing different trends
• Weights, fudge factors
• Find memory leak, if any
• Implement RU barrel shifter for LHC case
• See if it allows the DAQ to keep up with L1
• Use simple BCN
• Apply weights, fudge factors later