Go to the Indico Home Page Oldest event Older event Up to category Newer event Newest event Printable version
 
Focus on:  Hide Contributions
  • 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

    BLonD code development meeting

    → Europe/Zurich
    CERN

    CERN

    Video conferencing only
    BLonD code development meeting
    Zoom Meeting ID
    93916228605
    Host
    Helga Timko
    Useful links
    Join via phone
    Zoom URL
      • 15:00 → 15:10
        News 10m
        Speakers: Alexandre Lasheen (CERN), Giulia Papotti (CERN), Heiko Damerau (CERN), Ivan Karpov (CERN), Konstantinos Iliakis (CERN), Luis Eduardo Medina Medrano (CERN), Markus Schwarz (KIT, Germany), Panagiotis Tsapatsaris (NTUA), Simon Albright (CERN), Theodoros Argyropoulos (CERN)

        Welcome Kostis (generic optimization) and George (GPU optimization)

        Latest commits - SPS OTFB, bi-gaussian, bug fix for == vs is to be done

        Alex - discussions with Xiaofeng; invite to a BLonD meeting to present and discuss

        Fabian - muon collider BLonD works with multiple stations!, muon decay modelling -> short presentation next time

      • 15:10 → 16:00
        Memory limitations for HPC simulations and HTCondor BLonD library compilation 50m
        Speaker: Konstantinos Iliakis (CERN)
        H5Py For Extremely Large Scale Simulations.pdf

        Memory limitations for HPC simulations

        • Limited by the RAM of a single node
        • e.g. 2 TB of RAM available, but only 64 GB can be exploited
        • Solution: chunked data sets using H5Py
        • Using H5Py for multi-bunch distribution generation
          • Generate each bunch in an h5 dataset
          • Store h5 coordinates in a beam object and split beam
          • Same if distribution was read from a stored file
        • Follow-up: Alex, Kostis, Helga -> what to do with bunch generation routine
        • Q: usage of Spark - would solve the issue but too large overhead
        • Q: zar or dusk to offload arrays that are not really used by tracking, just available to user
          • Consider using new modules
        • Q: what is a good chunk size? affects performance, should be less than 1 M

        HTCondor runtimes

        • Cluster compilation issues
          • Traditionally only Intel servers, now also AMD
          • BLonD should be compiled for different vendors and models of architectures
          • Solution1: compile before running the simulation -- can crash for parameter scans
          • Solution2: compile for each machine -- not ideal for heterogeneous lxplus machine
          • Best: compile and load BLonD library depending on environment variable
            • To be implemented soon on the master branch

        LHC ramp simulation

        • Example of Viviana's heavy simulation
          • 14 M turns, 1 M particles, 2.6 M FFT points
          • dominated by induced voltage calculation
            • Reduce FFT size if possible
            • Or use GPU architecture