Skip to content

FTAG truth decoration fix: skip unstable or low pT particles

Samuel Van Stroud requested to merge (removed):svanstro/23.0-ftagTruthFix into 23.0

This is a follow up to !62544 (merged) which avoids decorating the entire truth particle collection, which is expensive and was observed to significantly degrade event processing time as discussed in ATLASG-2525.

I just added a simple check on the status and pt of the truth particle, which bring the performance back in line with the pre-MR efficiency to within a percent or two. Going from 500 MeV to 1 GeV does not noticably improve things further so the selection is left at 500 MeV for now.

23.0
Derivation 12:51:35 PerfMonMTSvc         INFO CPU usage per event [ms]:          735
Derivation 12:51:35 PerfMonMTSvc         INFO Events per second:                 1.314

500 MeV Selection
Derivation 13:14:34 PerfMonMTSvc         INFO CPU usage per event [ms]:          662
Derivation 13:14:34 PerfMonMTSvc         INFO Events per second:                 1.462

1 GeV Selection
Derivation 13:43:44 PerfMonMTSvc         INFO CPU usage per event [ms]:          664
Derivation 13:43:44 PerfMonMTSvc         INFO Events per second:                 1.446

If the small residual performance degredation is not acceptable, we can also think about only enabling the decorations for FTAG1, and not other formats.

cc @bdong @ligang

Edited by Samuel Van Stroud

Merge request reports