A study on classification of HPC jobs based on energy consumption

A study on classification of HPC jobs based on energy consumption

Wednesday, June 1, 2022 1:28 PM to 1:32 PM · 4 min. (Europe/Berlin)
Hall D - 2nd Floor

Information

CINES is a French national tier-1 supercomputing center for higher education. On one hand, its main assignments is to provide computing ressources for the academic research community and on the other hand, it offers facilities for permanent archiving of electronic data. CINES hosts Occigen, a 3.5 Pflop/s supercomputer enclosing Broadwell and Haswell processors.

Data on jobs running on Occigen is daily collected in a data lake and is used to carryout studies to gain valuable insights on the actual usage of the computational resources. In this study, we present some observations from the jobs’ energy consumption data and relevant parameters on overall machine, for the period June to August 2021.

We began by running three benchmark cases three times for one hour and collected their energy consumption in order to develop a classification criteria.

The results obtained highlight that a significant part of the jobs consumes energy at the level of idle jobs which is not optimal. This provides an opportunity of further studies and classifications on the different behaviors of communities and the use of computing resources.
Contributors:

  • Carla Andrieu (CINES)
  • Umesh Seth (CINES)
Format
On-site