AgML, CA, Stv review findings and action items

Summarizing the PSN0552 : STAR Tracking Components Review - Stv, CA and AgML from the You do not have access to view this node, we extract the key items to address below:


Issues - items in Italic are not important for moving to integration, items in red should be addressed before integration, items in blue are side actions needed before the follow-up review, items in green are minor and can be addressed after integration.

  • AgML - green light to move forward 2012/03/21

    • Finding: Pending issues of track losses and material discrepancies -> deploy.
      Attention: geometries > 2011 and beyond would use AgML (especially, 2012)

    • Can “cons” be speed up?
      Status 2012/02/22: First version of a stand-alone Makefile made. Uses trick of -jNCpu (bit not cons integrated)
      Status 2012/03/09: First integration to "cons" attempted. Discussed splitting the dictionary build (otherwise, the final step based on a large CINT takes many minutes)
      Status 2012/03/12: cons seem to compile in one pass but final step leads to duplicate link symbol (rule bug).
      Status 2012/03/24: Reshaped rules and generates all in one pass, dictionary still not done at first (will readdress later). General feel is that dictionary should be generated for individual files to avoid a final large dict co

    • Can one use NX? [FGT concerns]
      Status on 2012/03/06: Ready to be used. After some deployment issues with NX for general use, declared usable 2012/03/06 and users informed here. See also RT #2275.

    • Configuration problems with MTD and issues with PMD considered minor (but needs to be understood)

    • Eloss shift in EEMC and ESDM issues – is this understood?
      [Test suggested i.e. Cut in space or dE versus “x” suggsted ... other ideas floated around]
      Status on 2012/02/02: Jason promises to have an answer by Monday (will test with offset in material if consistent)
      Status on 2012/02/06: AgML hits (starsim) vs AgSTAR hits (starsim) in the ESMD explains the "from where can it come from"
      Status on 2012/03/14: Issue resolved re-explained at  You do not have access to view this node S&C meeting (slide 4). The SMD strips were not placed at the proper position due to a FORtran/compiler bug we do not understand (the equivalent of a global define is not doing the proper in-lining, replacing the expression by the value literal fixes the issue).

    • Edge material differences (end of the TPC)
      Status on 2012/02/02: May be a bug (devitaion from Geant3 expectations on treating “many”) in TGeo (??)
      Status on 2012/02/06: Extra medium definition added makes the
      AgML-->AgSTAR and AgSTAR comparison exact. Problem then is that ROOT Geometries introduces an offset (Jason still looking) [+].

      Status on 2012/02/21: Resolved. The Phi asymmetry was found to be due to the loading of erroneous geometry versions in the chain [+].
    • General status
      Status on 2012/03/14: All issues found as resolved and You do not have access to view this node. "cons" integration pending but not a show stopper.
      Status on 2012/03/20: PAC and committee chair contacted for further comments on closure
      Status on 2012/03/21: OK to go PAC+S&C - announced here.
    • Improvements:
      • cons should generate separate dict for each source (would speed up the process)
      • cons still requires two passes due to the generated dict (and missing rule)
      • For now, generated geometries were added to cvs - would remove later (removed 2012/03/22)
      • EcalGeo6 generates warning - need to cleanup (cleaned 2012/03/22)
      • Multiple codes in StarAgmlLib generates compilation warning - need to cleanup (fixed 2012/03/22)

  • Stv/Sti/CA

    • Finding: Critical importance for STAR – endorse the approach

    • Hit errors and Xi^2 not correct as seen during the review needs addressing right away BEFORE the beginning of Run 12 . Global DCA obviously wrong (truncated).
      Status 2012/02/02: Remember this was related to a cut somewhere and was fine in the event.root but not in the MuDST. Victor will re-check this. Was this related to the DCA plot from Feng?
      Status 2012/03/22: Reminded Victor of the truncated DCA distribution
      Status 2012/04/13: Reminded Victor of the DCA truncation, sent slide extracts

    • Documentation is poor to nonexistent – to be addressed along with development
      Status 2012/10/00: Checked again in September and no Ddcumentation found (doxygen)

    • Using CA did not improve efficiencies – why?
      Status 2012/02/02: None, Victor still looking to understand the differences. The issue may be related to the fact that CA is used in Stv+CA in a different way.

      Status 2012/03/22: Resolved. New stat and results presented based on a statistically significant sample (~ 1,000 events). Sti, Sti+CA, Stv and Stv+CA compared. Generally, the results shows Sv seem to be better in finding tracks though quality were not presented. This result at least shows that Stv+CA improves over Stv alone - the core reason was that the seeds were from the outer sectors only.
    • Vertex finder issues to be resolved before moving forward – ppVF counted on by Spin not working for now
      Status 2012/02/02: Tried to contact the VF team many times but no answer
      Status 2012/02/15: New Email reminder sent - if not answer, will have to assume
      Status 2012/03/16: Answer received after a re-launch (+ multiple-tasks from all directions started by executive; VF tasks and priorities not discussed with S&C)

    • Need to work with TOF and FGT
      Status 2012/01/00: FGT working from "eval" as components are being integrated.
      Status 2012/03/29: Tracking "shortcut" taken (work requested from top executive management / S&C learn of it indirectly) though target of a usable Stv version by June was confirmed as in-line with program by spokesperson office (same day check). TOF code reshaped to work with a Geant3-less awareness.

    • Speed is significantly different – profiling needs to be addressed as development goes
      Status 2012/04/12:
      To date, no attempts were made to profile code - task reminded (issue was that Stv was much slower than Sti)
      Status 2012/05/10: Speed beleived to be related to filling internal debugging histograms (but quantification not done / TBC).


  • CA

    • Finding: demonstrated significant improvement in the tracking efficiency; endorse the approach + suggest to move this to the HLT

    • Integration issues, TBB and Vc dependence especially, to sort before integration (PAC and S&C).
      Concern of external package dependencies.
      Proposed on 2011/12/28: Vc in OPTSTAR, no TBB - Yuri considered the action items not part of the findings to address but as show stoppers. Pointed to the review bullet for further reading and took finding a solution on myself.
      Status on 2012/02/29: Issues of dependencies in TBB also raised by other experiments at the Tracking workshop #3. Renewed discussion with Yuri - agreed to remove dependence and done on the status date.
      Status on 2012/03/08: Vc dependencies discussed in // - General plan is to have the package pushed for integration into ROOT. Facilitated discussions with the ROOT team.
      Status on 2012/03/14: Initial discussion took place with the main developers
      Status on 2012/03/29: Vc integration in ROOT CVS trunk has begun (no cmake guidance)
      Status on 2012/04/02: First version of vc in root/misc/vc - [Matthias/Jerome] test 1 reveals some integration issues
      Status on 2012/04/11: Discussed support in ROOT with previous team lead
      Status on 2012/04/12: Second version tested, compiler bugs detection prevents smooth integration

      Status on 2012/04/20: Third version sent to ROOT team for further testing
      Status on 2012/05/19: Vc confirmed to come up in v5-34-02 or 03 but not in the original relrease. Code being checked for intergation.
      Status on 2012/11/29: Jerome addressed again Vc in ROOT while @ CERN - plan is to deploy with the next releases in February (both 5.xx and 6.xx). Long term support secured with GSI.
    • CA code does not adhere to STAR coding standards, no documentation, no header describing what code does what
      Status 2011/12/28: Yuri does not want to address this issue. This will be moved to Ivan and team for consideration and merged into the "documentation" item below.
      Status: Merged with documentation action item.

    • Documentation should be consolidated by the time of the deployment
      Status 2011/11/28: Sent an Email to Ivan as per the need to document and implement a doxygen based documentation. Subject "Summary of findings and action items, CA" (all items summarized). Generally accepted, action thought to be discussed during the tracking workshop (see below).
      Status 2012/02/29: Discussed documentation at the ALICE/CBM/STAR tracking workshop (see this goals, slides and this presentation summarizing the action item, written in the context of KFParticles and under "common package", enthusiasm to write documentation was not to the highest). No move so far.
      Status on 2012/03/20: Agreed with Ivan to renew discussion on this front soon (pending Vc issue). General pointer that a student may help around CHEP 2012 coming at BNL
      Status on 2012/11/30: Documentation discussed during the 4th tracking workshop - still non-existent and not converging (requests for documentation comes from more than STAR at this stage)

    • CA noted to possibly be usable for the FGT but no demonstration made
      Status 2012/03/29: Considering this item, it is unclear if this could be done - task force dispersion comment made to the executive management.
      Status 2012/04/13: Contacted FGT team - existing FGT workforce not investigating CA method for the first phase "prove the FGT works"
      Status 2012/10/18: Jason prototype tracking shows trakcing is possible with FGT - need PointMaker from FGT team
      Status 2012/12/06: Jason pulled back from FGT tracking (need FGT to pick up the initial work)

    • Tracks passing over dead area (sector 20) not altered by gap - recommend studying ghost, fakes, split tracks
      Status on 2011/12/28: Page sent with general proof of principle. Post in trackrev HN done as trackrev2011/96 and trackrev2011/97.

    • Number of possible points in StiCA was off by 1. This needs explaining and fix before full deployment.
      Status: An explanation was sent around and the issue fixed - need to locate this proof again and validate.
      Status 2012/03/12: Changes in code for RT #2295 is confusing and need clarification. The first change (#if 1) caused a 7% drop in efficiency in Sti. A clearer statement for the committee is needed.