Developer's meeting

Wednesday 3 Oct 2018, 17:00 → 18:00 Europe/Zurich

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

EHN1 / data taking:

• patch panel installed by Giovanna & Enrico
• new BNL712 card arrived, intend to return the BNL711 card now in use to BNL
• publisher problems:
  • if WIBs (hard-)reset: can happen that no data arrive anymore (whether using data or emulation); need to reprogram
    • Needs to be followed up with Frans/Andrea (already known as "link-down problem" and in Andrea's to-do list)
  • more information on transmission issues obtained in a serendipitous way:
  • 712 (single-CPU system): were using same pinning as for the 711 (which was on a 2-CPU system)
  • result: 2 publisher threads were using the same CPU cores (one on hyperthread)
  • attempting once more to use 10 links, but still finding errors (CPU utilisation is not problematic)
  • tentative conclusion that this is the PCIe priority issue? Needs re-discussion with Kai
  • should attempt to reproduce with minimal setup
  • note the BNL development of a FELIX demonstrator using PCIe4 (see presentation); however, this by itself will not suffice, given that we need to increase our input bandwidth
• client side & operations:
  • wider trigger window (200 ms)  works(?)
  • reordering/compression integrated, but some issues during running [ post-meeting follow-up by Roland: now solved ]
• RC: issues when card dead

documentation:

• Roland agrees that "proper" documentation is needed
• see google doc link to get started again; need to re-evaluate what information we want to end up in this document

overlay backward compatibility:

• Milo will put in a control word + a version number (in the fragment header)
  • version number available in metadata -- we should be able to use that? (Phil & Kurt have been discussing this; Milo will follow up)

tentative proposal for next meeting: October 17

17:00 → 17:20 Progress at EHN1 and experience with data

17:20 → 17:40 documentation