Developer's meeting

Wednesday 28 Nov 2018, 17:00 → 18:00 Europe/Zurich

Present: Phil, Roland, Filiberto, Milo, Frank, Eric

Roland: new machine

• 12 cores
• not using on-board NIC but rather a Mellanox 3 card
• some issues running felixcore
  • also icc issue (reproducibly crashes system -- kernel panic), will stick to gcc (6.2)
• have managed to have 10 links working
• Intel experts: advise to use stream (utility for networking benchmarking): stream
  • confirms that memory throughput was limiting the old machines, and is not a limitation anymore with the new machine
• completing tests

Phil: hit finding

• networking and input data now available; can send FELIX frames over network
• code using AVX that expands raw frames; re-ordering modified to use collection plane wires only
• processing of re-ordered data implemented by time window
• discussion:
  • prioritise for optimising the hit finding algoritm
  • investigate throughput (verify e.g. that AVX registers are not a limitation)
  • also would be good to integrate w/ artdaq
  • revised re-ordering initially also retains induction plane information (for now this is thrown out afterwards), so should be straightforward to retain longer if we are to replicate the DUNE situation
  • not yet looking at actual PD-SP data; would be good to do so to ascertain that issues like stuck bits do not have a large negative impact

17:00 → 17:20 EHN1 matters

17:20 → 17:40 IEEE RT Proceedings discussion

17:40 → 18:00 hit finding update

Speaker: Philip Andrew Rodrigues (University of Oxford (GB))