Choose timezone
Your profile timezone:
13 sep 2019 phone conf https://indico.cern.ch/event/782638/ Connected: Michele Michelotto, Domenico Giordano, Chris Hollowell, Manfred Alef, Andrea Valassi, Benoit Roland, Andrea Sciaba', Jean Michel Barbet, Martina Javurkova Apologies from Martin Bly We will have a very important meeting during the next preGDB. The agenda is alredy online https://indico.cern.ch/event/739897/ Chris on the HEPSCORE It a simple python script that by default now will run a prototype of HEPSCORE19, a candidate for the next HEPIX/WLCG benchmark to replace the old HEPSPEC06 Discussion on the geometric mean and scaling factor. Chris will provide a formula to explain what he is doing. Michele suggest to show some example that different approach can give wrong results while this is more suitable. HEPSCORE will be restructured after next HEPiX meeting to build a dedicated python modulo and enable future maintainability and testing Presentation from Benoit Roland on the usage of fast benchmark to measure the CPU resources available for Atlas. Benoit is availabel to run similar studies with the new HEPSCORE on the setup he already benchmarked Martina Javurkova: Update on ATLAS workloads - Spread in the scores of ATLAS gen-bmk are quite high but can be improved by either processing more than 5 events (default), e.g. 500 events, or by changing skipEvents from “17060000” to "0" in Generate_tf.py. - ATLAS sim-bmk is okay. - Some issues in ATLAS digi-reco-bmk: not enough events in input data file for hard-scattering events (by now only 5 events) to be increased; wrong formula to calculate total score from 4 sub-scores will be fixed in next release; issue when enabling AthenaMP: Lorenzo fixed some issues and is testing; memory consumption (rss) exceeds 2 GB per thread (provided by default WLCG WN), possible fixes: run (and enforce) multi-process, skip pile-up