HEP-Workloads mid-cycle Sprint meeting

Friday 5 Apr 2019, 14:30 → 15:00 Europe/Zurich

Manfred Alef (Karlsruhe Institute of Technology (KIT)), Domenico Giordano (CERN), Michele Michelotto

Active Sprint: https://its.cern.ch/jira/secure/RapidBoard.jspa?rapidView=6519

14:30 → 15:00 Sprint mid-cycle meeting

Speakers: Domenico Giordano (CERN), Manfred Alef (Karlsruhe Institute of Technology (KIT)), Michele Michelotto

• BMK-76: Spread in repeated results of ATLAS sim workload:
  • Issue with parser?
    • Example:
      • JSON file reports score of 0.0477 and avg of 0.0030 when running 40 copies
        • 40*0.0030 =/= 0.0477
      • Running parser by hand results in (correct) score of 0.1185
  • Apart from that, low spread of around 2% in the ATLAS sim scores
• All workload runs have been repeated with Turbo Boost disabled in the BIOS
  • Same spread as before
  • Scores around 10...20% lower than before
  • Replaced date with version info in the table of results
  • See Twiki for details
• Comparing SL7 versus SL6 (using Singularity to run the images):
  • Systems under test: E5-2660 v3 @ 2.60GHz, 2 of 4 nodes in 2U chassis
    • Higher scores on SL6:
      • ALICE: ~15%
      • ATLAS
        • gen: almost no clear difference (overlaid by spread?)
        • sim: ~4%
      • CMS: ~10%
      • LHCb: ~2%
    • Contrasting to Jean-Michel's results at IN2P3
  • SL6 results added to Twiki