Skip to content

Jet code cleanup follow up

Dillon Scott Fitzgerald requested to merge dillon/jet_code_cleanup into qee_upgrade_24

This is a follow-up MR to resolve some outstanding discussions from !3242 (merged). This closes https://gitlab.cern.ch/LHCb-QEE/qee-hlt-run3/-/issues/34 and #744 (note that all info is contained in the former issue and the latter is a duplicate of the second checkbox and is thus already resolved). I will paste the checklist here for clarity.

@dzuliani and @lsestini, the third item will require your input. I am happy to have a discussion to clarify the issue. In short, there was an implementation to call TOPO tags in sprucing lines defined in lines/qee/vjets.py even though the line names indicated SV tags, and the situation for lines/qee/top_muon_elec.py is unclear (line name says Bjets, so either tag could apply, but the implementation was consistent with lines/qee/vjets.py). I preserved this implementation of calling TOPO tags when cleaning up the code in !3242 (merged), but I am not sure if it was a mistake to begin with. The line names and the tags we are calling should be consistent, so we should either switch the line names or change the tag being called from TOPO to SV. What do you propose?

IMPORTANT it was decided to switch the jet tags in lines/qee/vjets.py and lines/qee/top_muon_elec.py from TOPO to SV. This will likely affect the sprucing bandwidth -- to be tested!

This MR will fix two "bugs" and report b/w changes separately for each. The first is already mentioned above, switching tags from TOPO to SV in lines/qee/vjets.py and lines/qee/top_muon_elec.py. The second is related to the definition of the make_Trijets function, see !3353 (comment 7926934).

TOPO to SV b/w testing

The main change that is being tested here is !3353 (691e75bc). The source branch is evaluated with all commits up to !3353 (47517ea5).

The hlt2 full stream has the same bandwidth and rate for the source and target branches, as expected:

  • rate = 1000 kHz * ( 52 evts / 10,000 evts ) = 5.2 kHz
  • b/w = 1000 kHz * (6.7 MB / 10,000 ) = 0.67 GB/s

The sprucing bandwidths for the source and target branch are very slightly different due to the differing filesize (see discussion above):

  • dillon/jet_code_cleanup with TOPO -> SV tag fix (source branch)
    • b/w = 5.2 kHz * (2.5 MB / 52) = 0.25 GB/s
  • qee_upgrade_24 (target branch)
    • b/w = 5.2 kHz * (2.4 MB / 52) = 0.24 GB/s

Fix to make_Trijets bandwidth testing (all commits included)

Here are the results from the b/w testing including all changes (final commit !3353 (1ea6094f))

The hlt2 full stream has the same bandwidth and rate for the source and target branches, as expected:

  • rate = 1000 kHz * ( 52 evts / 10,000 evts ) = 5.2 kHz
  • b/w = 1000 kHz * (6.7 MB / 10,000 ) = 0.67 GB/s

The sprucing bandwidths for the source and target branch are also the same in this case, which makes me think the b/w changes reported above before the fix to make_Trijets were really due to some artifact of the snippets / compression. The commits with the changes from TOPO -> SV are included in this round of tests as well, and no change was observed between the source and target branch.

  • dillon/jet_code_cleanup with all commits (source branch)
    • b/w = 5.2 kHz * (2.4 MB / 52) = 0.24 GB/s
  • qee_upgrade_24 (target branch)
    • b/w = 5.2 kHz * (2.4 MB / 52) = 0.24 GB/s

No changes observed w.r.t. target branch with all commits included!

Edited by Dillon Scott Fitzgerald

Merge request reports