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

    HEPiX Benchmarking Working Group

    → Europe/Zurich
    31/S-023 (CERN)

    31/S-023

    CERN

    22
    Show room on map
    Manfred Alef (Karlsruhe Institute of Technology (KIT)), Domenico Giordano (CERN), Michele Michelotto
      • 14:00 → 14:10
        News 10m
        Speakers: Domenico Giordano (CERN), Manfred Alef (Karlsruhe Institute of Technology (KIT)), Michele Michelotto
        chep2019-bmk.txt

        Benchmarking GPUs

        Conferences/Meetings

        • pre-GDB on CERN site: Tuesday 8 Oct 2019, 09:30 → 18:00 Europe/Zurich
          • This is a proposal to gather all together in a physical room
          • evaluate availability via https://doodle.com/poll/6tfe76t96h6a26hi
        • HEPiX Autumn 2019 Workshop 14 Oct - 18 Oct
          • Target this conference for a major status report of our work
        • CHEP 2019
          • Abstract submission closes:         20 May
          • Submit an abstract for oral (see proposal from A. Valassi)
            • Please volunteer as speaker

         

        Sprint review reorganization

        • Proposal: move the bi-weekly review to the week that currently hosts the mid-term review meeting, in order to give more time to presentations in this meeting.
        • Dedicate just the spare time during this meeting to the discussion of standing issues 
        • Everybody agrees to make this change
        • Proposal: move the bi-weekly review to the week that cur

        Reproducibility of the results of the candidate benchmarks

        Chris H. reports on the differences amongst Docker on SL6, SL7 and Singularity. There are differences indeed but they are very small. 

        Jean Michel B. - Running CMS with varying number of events on CentOS7 and SL6 on singularity. Version 0.4 had a bug but wow he is running v0.5

        Using the tool "cpupower" to check for saturation of resources and monitor the core frequencies

        On the SL6.5 spread affected by some incomplete simulation leading to smaller step.root file. Removing the abnormal runs  givers stable results. In version 0.5 these problem should disappear. However there is also an issue that David Lange is investigating. This issue is very rare in the production so it's not a problem for them but could invalidate our results. About the usage of STDEV or the RANGE . We should use a set of script and uniform way to do it.  RANGE is MAX-MIN

        we would report both STDEV/AVERAGE and RANGE/AVERAGE

        Michele M 

        Michele - was running atlas gen, sim, and cms gen-sim . The version V0.5 fixes one problem with v0.4

        Domenico suggest to un atlas-gen-bmk -e 100 instead of the default -e 20

        Mandref A. 

        Checkpoint: The spread of the results and robustness

        • ATLAS GEN vey robust but spread very high
        • ATLAS SIM robust and stable but run time very long
        • CMS: robustness to be improved - spread almost ok - run time a bit long
        • ALICE: has a good run time but spread and robustness not ok
        • LHCB: robust and spread small
        • HS06: is robust, but it requires 32 bit package. The spread is usually less then 2%

        Martina

        Checked the linearity. Some problem due to the fact that she was running on a VM on which other benchmarks were running.

        Added a new parser to atlas-gen and atlas-sim

        Not clear the meaning of cpuError

        Martina is all investigating the use of prmon "PRocess MONitor" to monitor the rss consumption of a process and its children and evaluate the performance of HEP workflows

      • 14:10 → 14:25
        Validation of ATLAS workloads 15m
        Speaker: Martina Javurkova (Albert Ludwigs Universitaet Freiburg (DE))
        ValidationofATLASworkloads_Javurkova_100519.pdf
      • 14:25 → 14:45
        Reproducibility of Benchmark Results 20m
        Speakers: Manfred Alef (Karlsruhe Institute of Technology (KIT)), Mr Jean-Michel Barbet (Centre National de la Recherche Scientifique (FR)), Christopher Henry Hollowell (Brookhaven National Laboratory (US))
        atlas_gen_sim50.ods
        Reproducibility_of_HEP_workloads-2019-05-10.pdf
        Stability-Singularity-cms.pdf
      • 14:45 → 14:50
        Reproducibilty of Atlas and CMS workload 5m
        Speaker: Michele Michelotto (Università e INFN, Padova (IT))
        Reproducibilty-atlas-cms-wn012.pdf
        Reproducibilty-atlas-cms-wn012.pptx
      • 14:50 → 15:05
        Status/Closure of Sprint 3 15m
        Sprint3