Choose timezone
Your profile timezone:
Developer's meeting
→
Europe/Zurich
present: Milo, Roland, Enrico, Frank
EHN1 (Enrico, all):
- new Mellanox NICs installed in servers, but the cables are still lacking, so the cards have not been exercised yet
- Frank: the company indicated that they ought to be available by the end of the month (October), so would hope to have them soon
- tested with WIB (last Wednesday): took run with timing & FELIX included, but found mismatched timestamps
- for now, have developed 'final' BR
- stand-alone timing receiver, to which BR processes can 'subscribe' to receive event requests -- working
- will re-integrate into BR (but using a simplified communication of the event request from the event builder)
- plan is to have another test with WIB and timing soon, with the above fixes
- BR machines: not purchased yet, idea is to go ahead with other developments first before doing so
Data re-ordering / compression (Milo):
- (see slides)
- compression factor of 4 achieved when using "previous value subtracted" ADC values, after also re-ordering the data, and using the QAT deflate() implementation
- timing: 9600 frames "manipulated" in 8 ms (reduced to less than 3 ms when using 8 threads)
- Frank: will we have enough threads in the BR machines? Roland: when using hyperthreading, could expect up to 48 threads (this should certainly suffice)
- worry about slowness of data access (any ADC value retrieval requires accessing the whole waveform up to the requested sample)
- Frank: please verify (with https://docs.google.com/document/d/1ocGDkM4QcD0glo5BaSAa-1HlKWLXE4wir3AuCMPke3U/edit?ts=5967ce0f and/or by contacting Tom Junk) that this is an issue in practice
- Milo: will investigate more
Next discussion: planned tentatively for November 15 (with apologies to Roland, who may not be able to attend)
There are minutes attached to this event.
Show them.