Dortmund 2021 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 91: GRID computing

T 91.4: Vortrag

Donnerstag, 18. März 2021, 16:45–17:00, Tp

Performance monitoring of the opportunistic resource NEMO at ATLAS-BFG — Michael Böhler, Anton J. Gamel, •Stefan Kroboth, Benoit Roland, Benjamin Rottler, and Markus Schumacher — Albert-Ludwigs-Universität Freiburg

The workload of computing clusters is typically unpredictable and tends to alternate between over- and under-utilization of the available resources. The software COBalD/TARDIS provides an easy way to opportunistically make under-utilized resources of one cluster available to another cluster. Fine-tuning of the involved software infrastructure to optimize efficiency and user experience needs to be performed in a production environment and is therefore difficult without continuous monitoring of logs and meaningful metrics. In this work we present the current situation at Freiburg University where resources of the NEMO cluster are used to extend the WLCG-Tier-2/3 cluster ATLAS-BFG in an opportunistic fashion using COBalD/TARDIS. The talk covers the tools involved in the collection and analysis of logs and metrics acquired from different sources within the ATLAS-BFG and opportunistic NEMO. Examples of how the aggregation of logs and the monitoring of metrics aids decision-making are shown. Besides fine-tuning of the involved tools, this setup can also be used to detect problems and anomalies early on. It furthermore serves as a basis for the future development of an accounting system for compute infrastructure which involves opportunistically integrated resources.