Skip to content

HLTPFlow trk extrapolation: Loosening tolerance setting

Claire Antel requested to merge cantel/athena:23.0_trkXtrapolate_tightening into 23.0

Purpose

Loosening tolerance of STEP propagator used in the track extrapolation to calorimeter for PFlow reconstruction in the HLT in order to reduce CPU at the HLT (ATR-27539). Track extrapolation CPU vs performance optimisation was studied by @otheiner for his QT ATR-22226 (following suggestions in official ATR-21636 JIRA for HLT pflow optimisations general). The new setting was validated in HLT Reprocessing ATR-27324 where a 25% CPU reduction for HLTPFTrackExtension_ftf alone is measured with very minute changes to physics distributions (see discussions in JIRA and Ondrej's summary slides). It does affect trigger counts for a number of chains, but mostly by less than 1%. Impact should be on the trigger turn-ons only from slightly different reconstructed jet momenta - verified for some chains. HLT reprocessing was signed off by jet, met and b-jet.

Changes

  • Changing default tolerance of Trigger.Jet.PFlowTolerance ConfigFlag from 10-4 to 10-2 (one line change). Deciding to keep ConfigFlag for now (as opposed to hardcoding in python), in case of issues and as we could consider testing a further loosened tolerance.
  • Updating trigger count reference files.

Affects...

  • reduced CPU usage at HLT
  • Small change in physics - expected to be on trigger turn-ons from slightly different reconstructed jet momenta.
  • reference trigger counts

tagging @otheiner @miochoa @xingguo @lbeemste @lshi @jwuerzin @sidoti @sshaw @tamartin

Merge request reports