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
Videoconference Rooms
HEPiX_Benchmarking_Working_Group
Name
HEPiX_Benchmarking_Working_Group
Description
Semi-permanent room for the sessions of the HEPIX Benchmarking working group
Extension
10535458
Owner
Helge Meinhard
Auto-join URL
Useful links
Phone numbers
    • 14:00 14:10
      News 10m
      Speakers: Domenico Giordano (CERN), Manfred Alef (Karlsruhe Institute of Technology (KIT)), Michele Michelotto

      Pre-GDB:

      The Benchmarking pre-GDB has been rescheduled, because some participants are not available in April, or have desired more time for preparing their talks.

      NEW SCHEDULE: Tuesday 12 Jun 2018, probably starting at noon. Please, safe the date!

    • 14:10 14:20
      Spectre v2 microcode patches: impact on performance? 10m

      Meltdown and Spectre Security Patches:

      Intel has released new microcodes to fix the Spectre v2 security issue for a lot of processor architectures (Nehalem, Sandy Bridge, Ivy Bridge, Haswell, Broadwell, ...). Download the 'microcode-20180312.tgz' tarball from https://downloadcenter.intel.com/download/27431/Linux-Processor-Microcode-Data-File?v=t, unpack it into /lib/firmware, and reload the microcode (reboot is not required). Example (Sandy Bridge host, SL6):

      # ls /lib/firmware/intel-ucode/
      06-3f-02  06-4f-01  06-55-04
      # (cd /lib/firmware ; tar xzf /root/microcode-20180312.tgz)
      # ls /lib/firmware/intel-ucode/                         
      06-03-02  06-06-05  06-08-01  06-0a-01  06-0f-02  06-16-01  06-1c-02  06-26-01  06-3c-03  06-3f-04  06-55-03  06-5c-09  06-9e-0a  0f-02-05  0f-03-04  0f-04-09  06-05-00  06-06-0a  06-08-03  06-0b-01  06-0f-06  06-17-06  06-1c-0a  06-2a-07  06-3d-04  06-45-01  06-55-04  06-5e-03  06-9e-0b  0f-02-06  0f-04-01  0f-04-0a  06-05-01  06-06-0d  06-08-06  06-0b-04  06-0f-07  06-17-07  06-1d-01  06-2d-06  06-3e-04  06-46-01  06-56-02  06-7a-01  0f-00-07  0f-02-07  0f-04-03  0f-06-02  06-05-02  06-07-01  06-08-0a  06-0d-06  06-0f-0a  06-17-0a  06-1e-05  06-2d-07  06-3e-06  06-47-01  06-56-03  06-8e-09  0f-00-0a  0f-02-09  0f-04-04  0f-06-04  06-05-03  06-07-02  06-09-05  06-0e-08  06-0f-0b  06-1a-04  06-25-02  06-2f-02  06-3e-07  06-4e-03  06-56-04  06-8e-0a  0f-01-02  0f-03-02  0f-04-07  0f-06-05  06-06-00  06-07-03  06-0a-00  06-0e-0c  06-0f-0d  06-1a-05  06-25-05  06-3a-09  06-3f-02  06-4f-01  06-56-05  06-9e-09  0f-02-04  0f-03-03  0f-04-08  0f-06-08
      # grep microcode /proc/cpuinfo | uniq
      microcode       : 1808
      # echo 1 > /sys/devices/system/cpu/cpu0/microcode/reload
      # grep microcode /proc/cpuinfo | uniq
      microcode       : 1811
      #

      Remarks:

      1. If the final 'grep microcode /proc/cpuinfo | uniq' command prints the new release as well as the old one, then reload the new microcode to all cores, until only the new microcode revision appears in the output
      2. Warning: some systems at GridKa hanging after reload - run update on idle host if possible
      3. Software vendors (SL, SLC, CentOS) are not providing updated RPM files so far, because they are preferring Retpoline. However, there was rumor that Retpoline is not sufficient; therefore also microcode updates are still recommended. (You can also ask your silicon vendor for latest BIOS release implementing the microcode updates.)

      ACTION ITEM: Can site admins apply the microcode patches, rerun the benchmarks, and present results at one of the next meetings, please?

      • GridKa: no significant impact on performance found, fluctuations in the HS06 scores as usual
      • Other sites?
      • Reference workloads, and user jobs?
Your browser is out of date!

Update your browser to view this website correctly. Update my browser now

×