Dortmund 2021 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 91: GRID computing

T 91.2: Vortrag

Donnerstag, 18. März 2021, 16:15–16:30, Tp

Job Shaping with HammerCloud ATLAS — •Michael Böhler, David Hohn, and Markus Schumacher — Albert-Ludwigs-Universität, Freiburg, Deutschland

The functionality of the compute sites of the Worldwide LHC Computing Grid for the ATLAS and CMS experiments is verified by a large number of experiment specific test jobs. These jobs are steered, controlled and monitored by the HammerCloud testing infrastructure. HammerCloud ATLAS runs different functional tests, continuously checking the site status by representative MC simulation and analysis jobs. If these test jobs fail, the site is automatically excluded from central ATLAS job brokerage system: only test jobs will be send to the site until the test results succeed again. The auto-exclusion mechanism increases the success rate of the user jobs by only allowing job brokerage to healthy sites.

The aim of Job Shaping, which is discussed in this talk, is to speed up auto-exclude and re-include decisions made by HammerCloud. This is to be achieved by dynamically adjusting the frequency of test jobs based on latest test job results. Dedicated visualizations are developed to provide intelligible information. Additionally, specialized debug test jobs can be sent to problematic sites to identify root causes of problems like failing or missing test job results. The additional information of the debug jobs will provide more detailed data in order to help problem solving and identifying failure patterns. Therefore new test templates are developed which focus on testing specific components of the site functionality.