28th Conference on Computing in High Energy and Nuclear Physics (CHEP 2026)

Data-Driven Validation of HS23 with CMS Grid Job Monitoring Data

Not scheduled

18m

Chulalongkorn University

Poster Presentation Track 7 - Computing infrastructure and sustainability Poster

Speaker

Robin Hofsaess

Description

With this contribution, a data-driven method for the performance comparison of Grid sites is presented.
While the WLCG sites with an MoU typically report their performance in HS23, opportunistic sites, such as HPC or Tier-3 centers, usually don't.
For the comparison of opportunistically used HPC clusters in Germany, a method was developed to asses the performance of these sites based on CMS Production Job data, since they typically do not report verbose performance metrics and monitoring data due to security limitations.
At the same time, the method provides a robust, data-driven validation of the HEPScore benchmark, underlying its representativeness for HEP applications, as will be presented with this contribution.
Finally, the developed method also allows to identify deviations to the officially reported benchmark values, e.g., if hardware changed but benchmark results were not updated, in a sustainable way based only on already available data.

In this contribution, the method is described together with a performance study conducted in January 2025 based on CMS monitoring data.