Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 12: Data, AI, Computing, Electronics II (Data Management, Workflow)

T 12.4: Talk

Monday, March 31, 2025, 17:30–17:45, VG 2.102

Pre-cache tests with the WLCG Tier-2 centre GoeGrid and the NHR HPC cluster Emmy using workflows of the ATLAS collaboration at the LHC — •Inga Łakomiec, Saidev Polisetty, Arnulf Quadt, and Sebastian Wozniewski — II Physikalisches Institut, Georg-August-Universität Göttingen, Friedrich-Hund-Platz 1, 37077 Göttingen

The GoeGrid centre in Goettingen is one of the WLCG Tier-2 sites and contributes to the ATLAS job processing and data storage. The HPC cluster Emmy by the National High Performance Computing (NHR) has been successfully connected with GoeGrid and ATLAS jobs can be run on its resources. However, there is no large local mass storage at Emmy for the WLCG operations. Therefore, data for jobs that are processed at Emmy is currently provided by the GoeGrid storage. A transition of storage and computing resources to the Helmholtz Centres and NHR sites respectively from the university based Tier-2 centres is planned in Germany in the next years. Since some NHR clusters will serve as a big computing centres without a large local mass storage, there is a need to prepare proper caching solutions and validate them.

Small local storage can be available at Emmy for the WLCG tasks after the storage centralisation. The current ATLAS workflow management has been tested for Emmy computing resources together with the small local storage (pre-cache) instance at GoeGrid. Results will be presented for the different number of CPU cores used by heavier workloads exclusively or a mix of production jobs. Then, the transfer and deletion of data in terms of a small disc size will be shown.

Keywords: WLCG; ATLAS; NHR; caching; HPC