Benchmark WG Meeting

Tuesday 12 Feb 2008, 17:00 → 18:00 Europe/Zurich

31-1-030 (CERN)

Helge Meinhard (CERN-IT)

As agreed last week, let's meet today at 16:00 h UTC (17:00 h CET, 08:00 h West Coast). Connection details are given below. Those at CERN can alternatively pass by my office. Starting Time: Feb 12, 2008 at 5:00 PM, Europe/Zurich Duration: 1:00 To join both the web and audio conference (recommended), click here: https://audioconf.cern.ch/call/0188422 If you only want to participate in the audio conference, you may dial in with the following numbers: Call: +41227676000 (Main) Enter access code: 0188422

CPU Benchmark WG Meeting Notes, February 12, 2008


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


1. Comments on minutes from Jan 23

• no comments

2. Review of action items from last minutes.

• Alex has done some work to resolve the rate vs base discussion (more to follow in later section).
• Martin's Barcelona machine is too unstable to benchmark. He suggest waiting for the third generation of barcelona CPUs.
• Experiment contacts have been using the Wiki and the mailing list to co-ordinate bechmarking. Franco is having difficulties with the mailing list. Helge will inquire.
• Now using indico for agenda and minutes.

3. Review of experiment progress

• Helge proposed that experiments submit a figure of merit which is positively increasing with machine performance. For example events per second.
• Peter Hristov asked if the same random seed should be used on all cores. This was discussed but the feeling was that if the number of events was large enough (100+ for example) then the random seed would be irrelevant. However this is important to check.
  

LHCb (Hubert Degaudenzi):

• working to find a reproducible estimator
• he has some results but they are not reproducible at the moment.

Alice (Peter Hristov):

• Asked question about random seeds (see above)

Atlas (Franco Brasolin):

• Has results which are posted to the wiki.

CMS (Gabriele Benelli):

• Gabriele had another meeting that he had to attend
• Peter Wegner is ready for Gabriele to test his system.

4. SPEC Rate vs SPEC Speed

• Alex investigated the SPEC code and had some interaction with Cloyce Spradling of SPEC. It turns out that the threads of SPEC rate are not specifically synchronized but do suffer "de-facto synchronization" as a result of the SPEC implementation. (see his mail [requires password] on the list archives).
• Therefore the priority should be on using multiple spec base, and collect spec rate data where ever possible.
• Thanks to Michele for pushing the issue so we discovered that SPEC rate is not explicitly synchronized.
• Michele noted that it's important that we note the distinction of the multiple spec base runs when explaining to vendors and showing out results.

6. Action items

There weren't too many well outlined action items from this meeting. Most work is in progress.

• Gabriele to try out DESY system.
• Further work by all experiments.

5. Next meeting

• Experiment checkpoint meeting Feb 20th, 16:00 UTC
• Planing meeting Feb 27, 16:00 UTC