Benchmark WG Meeting

Wednesday 2 Apr 2008, 17:00 → 18:00 Europe/Zurich

31-1-030 (CERN)

To join both the web and audio conference (recommended), click here: https://audioconf.cern.ch/call/0145273 If you only want to participate in the audio conference, you may dial in with the following numbers: Call: +41227676000 (Main) Enter access code: 0145273

CPU Benchmark WG Meeting Notes, April 2, 2008


Present:
Alex Iribarren, Helge Meinhard, CERN/IT
Franco Brasolin, CERN/ATLAS
Hubert Degaudenzi, CERN/LHCb
Manfred Alef, FZK Karlsruhe
Peter Hristov CERN/ALICE,
Michele Michelotto, INFN Padova
Martin Bly, RAL
Ian Gable, U Victoria
Sandy Philpot, JLab

1. Comments on minutes from March 19

• no comments

2. Review of action items

• Ian in progress with HTML posting
• Peter Hristov has resolved wiki loging and posted result (more later)
• Hubert is now running on lxbench machines

3. Progress from experiments.

Peter Hristov (ALICE)

• Now posted results to the wiki. Will make some messurements of reconstruction next week.
  
• He noted that there was little difference in the running times on the cores as expected.
• He plans on including nicer graphs into the excel files, but will have root graphs up shortly.

Franco Brasolin (ATLAS)

• planning another run on lxbench with ATLAS code release 14
• Also planning on running on the Zeuthen 2.83 GHz Harpertown (strongly encouraged by Helge as this is the highest performance Harpertown available within the group).

Gabriele Benelli (CMS)

• Not available due to having to give another CMS presentation

Hubert Degaudenzi (LHCb)

• Has run on all lxbench machines and encountered some thread issues on lxbench03 and 06 (which happen to be the Opterons)

3. AOB

• Helge received comments from one member of the LCG Management Board (Jeff Templon) that it would be better not to normalize to the performance of lxbench01 because it suffers from the 1GB/core memory limitation. Helge suggest normalizing to lxbench02.
  
• Andreas Hirstius has run the SPEC benchmarks on machines instrumented with perfmon, as well as instrumenting regular lxplus workernodes.
• Michele has made a Barcelona and a Hapertown available to Franco and Gabriele, and is also ready to make them available to LHCB and ALICE. There was some debate about the presence of the TLB bug on these machines.
  
• Martin has a 2.83GHz Happertown available if there is demand

4. Action items

• Martin to forward to the list what machines he has available.
• Ian to follow up on HTML posting again.
• Experiments encouraged to run on 2.83 GHz Hapertown available at Zeuthen.

5. Next meeting

• Next meeting will be held April 8, 16:00 UTC.