Choose timezone
Your profile timezone:
Chris Hollowell, Domenico Giordano, Manfred Alef, Martin Bly, Michele Michelotto, Olga Datskova, Luca Atzori, Viktor Khristenko , Martina Javurkova, Servesh Muralidharan
GDB in October dedicated to benchmakring , it will be announced at the next GDB. We plan to have a consolidated benchmark by that date
Talk submitted at CHEP 19 by Andrea Valassi on behalf of the HEPiX Benchmarking WG. At the moment only the 3 chairperson and Andrea are mentioned but the final talk and the article in the proceeding will be signed by all the people that have contributed or will do.
Olga show how she manager to run benchmark on the kubernetes cluster at CERN and how can she saturate resources. This is interesting since this can mimic a realistic production environment. WE can run a single benchmark (e.g. atlas-sim) and saturate several machine but also run in a mixed. We can for example see if it is better to run half simulation and half reco.
We can mimic a chaotic system very close to a realistic environment.
Is this scalable? This is interesting for us because sometimes our benchmarks fail but this is rare and difficult and lengthy to reproduce.
This can be useful also for the WLCG cost model wg
Viktor working on DEEP-EST project (see slides)
Exploiting parallel resources in HPC/GPU for CMS reco. In CUDA or AMD HIP
Viktor can provide a clean VM with Nvidia V100 + docker-ce + nvidia-docker2
Provide a docker file , encapsulate open data (can't use regular cms events)
This can be useful to start running a workload on different GPU and noGPU resources.
Next meeting in two weeks. Domenico propose to configure an automatic announcement