MicroBooNE Discussion 15Oct2010

Eric has resurrected the old miniboone node.
Thinking of throughput tests to perform using the old code.

10 SEB PCs (sub-event buffer)
uBDPC assembler

1000 channels * 2 bytes per = 2000 bytes per sampling cycle

Need machines: SEB, uBDPC. 3PCIe slots, 3 hard drives slots (continuous read/write, RAID) 2U server

to disk: 50MB/s
PCIe cards: 2 fiber links out of this to the electronics - 50MB/s in
ethernet out: 5MB/s

assembler in: 5 * 10 SEBs = 50MB/s

1.6ms windows sampled at 2MHz coming in (1.6ms buffer)
3 windows are buffered and send - previous

once a week build a supernova event - go to ten drives, 1TB events
180GB * 10 = 1.8TB

transfer out needs to be done concurrently with writing

My calculations

2M samples * 1000 channels * 2 bytes/channel * 1.6ms/window = 6.4MB / window * 15Hz window rate = 97MB/s per node

Final thoughts...

I had to leave early (jbk). Ron and Gennadiy continued discussing the problem.

What is interesting about this problem?

  • good learning system
  • good test ground for DAQ-in-a-box?
  • good test ground for compact DAQ system? (with redundancy)