==================================================
Minutes of the Tracking Meeting, November 28, 2007
==================================================
taken by U. Husemann


Juan Pablo Fernandez: ISL Alignmen in 7.0.0pre2
===============================================

- Dataset: jpmm0j (Period 11)
- Testing default (pass 17) alignment table with Gen7 pre-release
- Track selection: pt, d0, Silicon hit/strip isolation
  o old: all Gen6 DefTracks
  o new: separate OI, IO, Backward
- Unbiased residuals for example central barrel: RMS 150-250 microns
- Example forward barrel: dominated by backward tracks, 
  all COT based tracks systematically shifted by 100 microns (misalignment?)
- Layer 6 central: all residuals within 30 microns
- Layer 6 forward/backward: residuals within 40 microns
- Layer 7: smaller residuals on inner bulkheads, fairly large for outer BH
- Suggestion: plot residuals as function of z position of track intercept
- Expected impact on physics: mainly resolution, rather than efficiency
- Discussion: 
  o Why do OI, IO, Backward residuals look so different?
  o Bias due to algorithm (different resolutions, numbers of hits, ...)?


Ulrich Husemann: Update on Gen7 Timing
======================================

- Last collaboration meeting: 
  o Testing ideas for optimizations to get to 1.5 times Gen6
  o Request: weight CPU times with luminosity profile
- Long list of algorithm optimizations implemented since 
  (see minutes of Nov 1 meeting)
- Code optimization: small effect in maxopt executables
- Luminosity profile:
  o Fit number of unique CSL events vs. lumi for 85 runs over 200e30
    -> shape approx. universal (power law)
  o Convolute lumi profile with parameterization of CPU time vs. lumi
    -> average CPU time per event
- Results:
  o Gen7-pre2 was 2.5 times slower than Gen6
  o Gen7-pre3: 1.5 times Gen6 (keeping promise from collaboration meeting)
  o Further tests with FWD seed pT cut of 4 GeV, COT and backward segment
    cut of 0.5 GeV -> each around 1.25 times Gen6
  o Request: softer cuts on both (FWD: 2 GeV, segments: 0.4 GeV) 
    -> done after the meeting: 1.2 times Gen6
- Breakdown per module:
  o Aseet will investigate CT_TrackingModule performance (>2 times Gen6)
  o Antonio expects improvement from updated BackwardTracking
- Detailed tests with profiler: no silver bullet, 
  there may be a few small improvements


All: Discussion on Gen7
=======================

- Pending issues for next pre-release:
  o Aseet: check CT_TrackingModule
  o Antonio: finish changes to BackwardTracking
    (addressing issues with resolution, COT hit attachment, overall cleanup)
- Possible way of showing improvements: getting more tracks of "good" quality
  (which may be hard to define generally)
- Doug's suggestion: present current status at next week's JP meeting,
  second presentation with detailed validation later