Choose timezone
Your profile timezone:
Developer's meeting
→
Europe/Zurich
Present: Giovanna, Enrico, Junbin, Kai, Eric, Milo, Frank
EHN1 (Enrico):
- previously: problem w/ branch used by Frans, related to setup; now fixed
- multi-descriptor builds now working: 6--8 Full-mode links (out of max 12 DMA descriptors) on one BNL711 card working (with fdaq and connected to WIB)
- actually 11 links with most recent f/w build; tested one by one, so remaining DMA transfers are "lost" (but still useful as they do imply PCIe traffic)
- means 4--6 links are not working; some do work with internal emulation and only fail for WIB data, but in most cases also internal emulation failed
- so in all likelihood, there is more than a single issue to deal with
- next: test also with felixcore instead of fdaq
- data integrity: issues with data as retrieved from data logger
- should investigate whether this is still the case (in principle the event building and storage should not affect the data format at all)
- can also check with RCE data
WIB f/w (Junbin):
- discovered that issues with CRC errors and data losses were due to the EventBuilder f/w block
- fixed using (substantial?) f/w modification, borrowed from FELIX front-end design
- different clocking, extra FIFOs to convert from 240 MHz to 120 MHz clock
- recent WIB f/w versions (18032901, 18041701) are unable to use a local clock (even if they can use the PDTS clock)
- similar for RCE, as confirmed by Justin
- will attempt to get real timing system from Matt
- would be good to do a longer test also again at CERN (last time this was done is a while ago)
- this would use the 18032901 f/w version?
compression (Milo):
- looking at μBooNE MC (RMS noise values between 20 and 100 ADC counts -- seems very large?)
- upshot from software compression: good compression results but too slow by an order of magnitude (~ 200 ms per fragment)
- QAT on the same data: compression up to a factor of (just under) 4
- observation: re-ordering data is important, but previous-value subtraction appears to have very little effect
- next:
- check QAT results as a function of RMS noise
- check also on μBooNE data
- but note (Eric): the data events used here suffered from very high coherent noise, and may need some pre-processing before they are useful
There are minutes attached to this event.
Show them.