Aachen 2019 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Machine-learning methods and computing in astroparticle physics

AKPIK 3.9: Talk

Wednesday, March 27, 2019, 17:20–17:30, H06

Benchmarking of compute resources — •Benoit Roland, Felix Buehrer, Anton Gamel, and Markus Schumacher — Albert Ludwigs Universitaet Freiburg, Freiburg, Germany

The high luminosity phase of the LHC will be accompanied by a con- siderable increase in the need of compute resources and storage. In addition to dedicated GRID-sites, opportunistic resources can be used to fulfill the increasing requirements. These resources are heteroge- neous and their rapid classification is useful for an efficient utilization. The use of fast benchmarks as representative of CPU workload for GRID activity is presented within this context, with the aim to provide site monitoring and prompt matching between workload and available compute resources. The performance of several fast benchmarks have been studied and compared to the one of the legacy HEP-SPEC06 suite. The results have been obtained for different system configura- tions, with a particular focus on the behavior of the benchmarks when hyperthreading is active and their performance loss when virtualiza- tion is used.