Choose timezone
Your profile timezone:
Developer's meeting
→
Europe/Zurich
Present: Brett, Junbin, Eric, Enrico, Roland, Milo, Frank
EHN1:
- new f/w from Frans with multiple descriptors, tested in depth
- but unable to run multiple fdaq instances (in contrast to Frans), or (alternatively) felixcore
- leads to FELIX host crashes; Giovanna has sent system log files to Frans, to help debugging
- 2-descriptor f/w works stably (but this is simply the original DMA implementation)
- but unable to run multiple fdaq instances (in contrast to Frans), or (alternatively) felixcore
- trigger matching working with rate up to 24 Hz; back-pressure for higher rate
- Giovanna verified that even with emulated data this occurs, so it appears to be an issue with the raw data rate
- NB RCE are having similar problems and are limited to 6 Hz
- Enrico found an error with the CRC20 polynomial (which was also used by Dan), which could lead to spurious CRC errors
- thought that this could explain some of the issues seen at BNL as well
- plans:
- Enrico intends to continue on the topic of the "jumbo blocks", starting from the "main" branch rather than the multi-DMA one
- ideally, both features are to be used; but the expectation is that it will not be a large amount of work to port the jumbo blocks, once they are working
- in the Vertical Slice Test setup, 4 FEMBs should be available for FELIX
- Giovanna has been working on code to get the the trigger from the timing system instead of the Event Builder (i.e., back to the original plan)
- aim is to reduce latency
- Enrico intends to continue on the topic of the "jumbo blocks", starting from the "main" branch rather than the multi-DMA one
BNL:
- seeing corrupted data also when taking real data
- also still seeing link instability
- working on own version of WIB f/w, think this may be better protected against data corruption
- (Enrico:) this has not actually been tested at EHN1 yet
Discussions:
- (Frank:) Jos has indicated that he is continuing his studies of a single-DMA-descriptor solution with scatter/gather to distribute data over multiple subscribers (and with an additional throttling mechanism to ensure that different threads remain sufficiently synchronised)
- a potential issue would be the splitting of chunks over blocks; but if jumbo blocks could be used perhaps this should not be a serious issue
- Roland is interested in following up with Jos more (even if not immediately)
- we spent a bit of time discussing some aspects of the work towards DUNE
- merits of having fewer, but more powerful PCs (saving on Board Reader machines in favour of doing more on the FELIX host)
- no clear conclusion
- hardware path of QAT compression (on new hosts with on-board QAT FPGA): data returned to CPU memory after compression
Tentative agreement to meet next on June 20 (regular bi-weekly meeting slot)
There are minutes attached to this event.
Show them.