Participants: Alexey, Alla, Andreas Mathis, David, Dennis, Dmytro, Florian, Jens, Laurent, Markus, Mohammad, Radoslaw, Ruben, Sandro, Benedikt, Francesco, Jouri, Tom, Chiacomo

• MCH simulation (Laurent)
  • start simple and then refine
  • geometry in O2 now
  • hits in the detector are there, simple hits
  • currently in the works
    • geo fixes that are known to be necessary
    • add electronic chips (not in AliRoot as well)
    • uniform level of details (station1 very detailed, station2 basic)
    • radiography of detector elements (x/x0, material budget)
  • the hit map shown, position not correctly recorded so far, easy fix
  • next in the works
    • MID geometry and hits, not started at all
  • long term
    • digitiser
    • improve hits
    • both are linked, test beam to get updated information (not before autumn)
  • misc
    • strange bucket dependencies 
    • tree view of volumes, could not find in TGeoManager, done as text viewer
  • Q&A
    • is MID effort comparable?
      • probably less work, only one kind of detector
    • digitisation will stat in autumn, can we have an empty box for that?
      • could be done to have complete chain, just read hits and do something basic
    • bucket, couplings should be much less
    • tree printer of geo, nice to put in different level, move it somewhere

• StepLogger analysis framework, simulation optimisation (Benedikt)
  • cuts/processes via MaterialManager
  • first implementation for HALL and TPC
    • HALL ported from AliRoot  (comments?)
    • TPC taken from former commit (check, comments?)
  • continuous detector simulation QA
    • MCStepLogger
    • several examples of what can be monitored
      • steps, volume name, ...
    • bug with G4 discovered
      • neutrinos get stuck
      • G3 skip neutrinos by default
      • G4 does not, must be disabled
      • but possible issue with geometry, to be investigated before we skip neutrinos
    • add analysis on the fly
      • similar to RIVET approach
    • could have automatic analysis for each PR
  • Q&A
    • what is the purpose of redefinition of way cuts are applied?
      • to be able to have cuts applied on media level, independently on each media
      • so far was not there in O2, in AliRoot there is the big file
    •  

• New sensitive detector interface (Ivana)
  • introduce concept of sensitive detector in VMC
  • new interface class TVirtualMCSensitiveDetector
    • with pure virtual function to be implemented by user
  • details on how to use the interface to set/access
  • implemented for GEANT3/4 (no gain in G3 expected, retrieved from std::map), Fluka to be done
  • examples provided in VMC E03 into two versions, both included in test suites 
  • Q&A
    • if we move to this framework Fluka won't work for the moment
      • implementation in Fluka is very fast
    • another thing that could not run is the MCStep
• Updates for MT (Ivana)
  • updated FairRoot and O2
    • MT not working anymore, was working few months ago
  • G4 and G4MT tests were disabled (creating noise)
    • putting these tests back is important 
  • Sandro: Comment on G4-MT performance: A possible slowdown (just by recompiling) was found and is followed up in JIRA https://alice.its.cern.ch/jira/browse/AOCSF-156
    • should be understood/investigated

• Update + status of DPL (Sandro)
  • reminder of current way of hit processing (FairRunAna)
  • move from FairRunAna to new layer (collaborative system that communicates via messages)
    • move from sequential to more flexible, configurable with potential of parallelism
    • better model with actual data flow
  • prototype developed targeting TPC digitisation
  • main ideas and components outlined
  • demonstrations
    • parallelism
    • background-signal merging
  • one caveat
  • next steps 
  • Q&A
    • Ivana: if you run MT, simulation is parallel and not sequential which could direclty be exploited for digitzation?
    • did you take into account this possibility?
      • Sandro: It's currently a different level of parallelism that is exploited in DPL which does not match having multiple files produced by MT
      • Ivana encouraged to make a sketch of her idea
    • Laurent: why are hits not send as input as this would allow processing across one node and potentially direct forwarding from simulation to digitization?
      • Sandro: agree. Idea already tried out (see talk in Wuhan Dez 2017). However, for the DPL example, the focus was on parallelism and sharing the collision setup. Sending hits as input foreseen in the next iteration (although we have to see how to it efficiently). Currently no support/experience with going across node in DPL.

• AOB
  • Ruben, implemented simulation of QED electrons in ITS and MFT (AliRoot)
    • port old PR of Roberto