Go to the Indico Home Page Oldest event Older event Up to category Newer event Newest event Printable version
  • Compact style
  • Indico style
  • Indico style - inline minutes
  • Indico style - numbered
  • Indico style - numbered + minutes
  • Indico Weeks View
    • Switch to the management area of this event
    Login

    EP R&D Software Working Group Meeting

    → Europe/Zurich
    32/1-A24 (CERN)

    32/1-A24

    CERN

    40
    Show room on map
    Live Notes
    EP R&D Software Working Group Meeting
    Zoom Meeting ID
    62893115044
    Host
    Graeme A Stewart
    Alternative host
    Andre Sailer
    Useful links
    Join via phone
    Zoom URL
    Hide

    Software R&D Working Meeting Minutes

    2022-04-13

    Apologies: Andre

    Room: Graeme, Placido, Witek, Paul, Erica, Valentin, Anna, Dalila

    Remote: Marco, Andi, Albetro, Benedikt, Brieuc, Gilherme, Jakob, Javier

    Introduction

    Re. resources in IT, we stress that this has moved beyond R&D, to production support.

    Hardware could be via cloud, we are disinterested in this!

    Important that resources are available for as long as the projects need them.

    Want to continue the arrangement on hardware hosting, so we check this.

    CLUE in Calorimetry

    • Q. CaloHits meed to be in Gaudi -> EDM4hep issue or k4CLUE?

    • A. Additional data can only be transient at the moment

      • Need to know if we have seeds/outliers, etc. for validation, which is extra data

    • What about the EDM4hep cluster collection?

      • It doesn’t have these additional fields that are needed for validation

    • Should follow up in EDM4hep meetings

      • Current model is simple, which reduces costs of implementation etc.
      • But could we allow for extensions/user data to cover this kind of use case?

     

    • Q. Andi: To make CLUE reusable and standalone, there us a tension between generic descriptions and specific detector issues, like calibrations.

    • A. Erica: Solution so far has been to duplicate, with specific CMS pieces in CMSSW. Would like to avoid that, and to go to just RecHit positions. In CMS there are different detector technologies that imply different cuts, but we believe we can handle this outside the core algorithm.

    • Andi, did this sort of thing at first in ACTS with templates, but that became very painful with compile times exploding.

    • Erica, at least we only really need RecHits for the core algorithm to function

     

    • Q. Paul: slide 4 had zero pileup plots and single events. How about with busier events?

    • A. Erica/Marco: Initial studies (no public plots yet) are quite promising in reconstruction perforamnce; converted photon reconstruction is not degraded at high energies. Also the CPU performance is good, overall time is 3-5s at 200 pile-up. This is 5% of the overall HL-LHC CMS reco.

     

    • Q. Anna: In k4CLUE which detectors are being used?

    • A. Erica: Using CLIC reconstrution, with EDM4hep data files and plain point inputs. It’s not at all detector specific.

     

    • Q. Placido, have you tested CPU and GPU versions to check for performance and stability?

    • A. Erica/Marco: Code is not yet single souce.

    • GPU gains x10 w.r.t. single-core; there is a multi-core comparison in the CLUE paper. Did use Kupla (from CASUS) to convert from NVidia code to CPU. Now moving towatds using Alpaka, so can support any backend that Alpaka can target. This includes memory managemnt (Andrea Bocci has improved this recently). Performance is as good as native CUDA code.

    • Placido - we need to test how to support this in k4hep, multi-platform issues.

     

    • Q. Andi: Understand that the inputs are very smple objects. What about maths librares, like Eigen?

    • A. Erica/Marco: There is a plain point-struct internally, not dependent on external libraries.

    • ACTS now using ‘algebra plugins’, that allow us to change backends.

    • CMSSW has a similar capability, but not yet available in the stand-alone version.

    • We have always held back by lack of testability in real scenarios for exploring this, which we want to overcome (cf. ATLAS LS1 synthetic tests that resulted in choice of Eigen)

     

    • Q. Bireuc: Will you port CLUE3D to Key4hep?

    • A. Erica: Yes, by making it part of the CLUE package.

     

    There are minutes attached to this event. Show them.