HEPiX Benchmarking Working Group

31/S-023 (CERN)



Show room on map
Manfred Alef (Karlsruhe Institute of Technology (KIT)), Domenico Giordano (CERN), Michele Michelotto

Meeting Date: 2/6/2017
Attendees:  D. Abdurachmanov, M. Alef,  L. Atzori, J-M. Barbet, A. Brosa,  A. Forti, D. Giordano,  C. Grigoras, M. Michelotto, A. Valassi

1) News

  • No objection to have an additional meeting on the 9th of June, in case there are updates to discuss in view of the WLCG workshop

2) Correlation of Benchmark Results, Job Performance, and the Number of Job Slots per Core (Manfred)

  • Manfred presents an update of his study, based on an indirect measurement of the job performance, using "top" as monitoring tool.  
    • Limitations are
      • Quick-and-dirty assessment, doesn’t take into account the job type (simulation, reconstruction, ...), or whether the jobs are part of the same production or not
  • Overall comment:
    • The measurement approach needs to be refined, there are several ways to better measure the cpu time of the jobs, gathering feedback from the Experiment job monitoring systems. 
      • Costin offers support to collect job information from Alice monitoring
      • Andrea S. can do the same for the Atlas and CMS jobs (already discussed)
  • Discussion about the current results
    • Results about Alice are in contradiction respect to what presented by Alice (Costin) in February 
      • Now it seems there is agreement with DB12cpp, whereas in February there was agreement with DB12
        • Costin offers to run again the tests using DB12cpp. He needs information about how to run it
          • Domenico will circulate information
    • Results about LHCb are completely off. Manfred is investigating further.
  • Actions:
    • Manfred will refine the study, with input from monitoring data of the experiments 
      • Costin will provide support for Alice, Sciaba' can help in extracting data for CMS and Atlas
      • Objective: to give an update asap, possibly before hte WLCG workshop, in order to be integrated in the WLCG report.
    • Costin will run the tests performed last February, integrating DB12cpp (documentation from Domenico)

3) AOB

  • Domenico gives a brief update about the task of including experiment workloads in containers.
    • CMS TTbar GEN-SIM is now also available for test. 
      • Still limited by the fact Frontier is needed. Domenico made a request to CMS DB (Giacomo Govi) to have an sqlite file. Waiting for a fix to the application that should dump a global tag into sqlite
    • Information about how to run the workload will be circulated soon, after having received feedback from CMS folks that are testing it (Pepe, David A., David L.) 
There are minutes attached to this event. Show them.
    • 2:00 PM 2:10 PM
      News 10m
      Speakers: Domenico Giordano (CERN), Manfred Alef (Karlsruhe Institute of Technology (KIT)), Michele Michelotto
      • WLCG Workshop
        • Benchmarking session moved to Wed 21 morning (11:00 to 12:30)
          • No CPU Unit session. It will be described in few slides in the introduction talk  
      • Coming WG meetings
        • 9 Jun: additional meeting in unusual week to gather further results for the workshop. 
          • to be confirmed  
        • 16 Jun: review of presentations / discussion session material 
    • 2:10 PM 2:30 PM
      Correlation of Benchmark Results, Job Performance, and the Number of Job Slots per Core 20m
      Speaker: Manfred Alef (Karlsruhe Institute of Technology (KIT))