============================================
Minutes of Tracking Meeting, August 29, 2007
============================================

taken by U. Husemann


Antonio Boveia: Processing Times for int10
==========================================

- Known differences between 6.1 production and 6.1.4int10
  o New CT_TrackingModule: expected to be 40% faster
    (before some of the changes)
  o New "backward tracking"
  o New forward silicon standalone
- Zee from 0h: 4.6 seconds (2 seconds from silicon standalone)
- High-pt muons from 0j era: 66 seconds (36 seconds from silicon standalone)
- Fair comparison on the CAF on the same min bias events (~100 random files)
  o Rough: mean luminosity per file
  o Detailed: compare mean times binned in 25e30 luminosity bins
- Offline group used parabolic fit to processing time vs. luminosity:
  o 6.1.4int10 takes factor of 2 longer than 6.1 production
  o New CT_TrackingModule - same time, maybe slightly longer at high lumi
  o New SISA - factor of 2 langer above 50e30
- First attempts to improve SISA processing time:
  o Cut at pT>10 GeV helps, but is a harsh cut
  o Code seems to have room for algorithm optimization
- Possible optimizations:
  o CT_TrackingModule & backward tracking: tune segment finding (Aseet?)
  o SISA: room for algorithm optimization (can the authors work on this?)
- Offline group's statement about computing time:
  o Prediction (parabolic) has some contingency, modest increase is ok 
  o Longer processing times: need to cut CAF time from users
  o Fair comparison: fold in processing time gain from not running track refit


Ji-Eun Jung: Beam Lines in Long Runs
===================================

- Compare old and new average beam lines for 231241, 236653, 231334
  o Averages are essentially unchanged
- Requests for checks:
  o Are first 60 segments identical between old and new?
  o Plot beam lines vs. segment number
  o Plot d0 vs. phi for the new beam lines (ask DQM group)
- Changes are not yet in the database, Bo suggests to check them in
  (no risk unless they are back-patched into calibration)
- Q: what if there is a segment with a missing beam line
  A: Take run average (=first entry)?