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

T: Fachverband Teilchenphysik

T 59: Grid-Computing und Software

T 59.7: Vortrag

Mittwoch, 27. März 2019, 17:30–17:45, S11

HEP Analyses in a Cloud — •Matthias J. Schnepf, R. Florian von Cube, Manuel Giffels, Christoph Heidecker, Günter Quast, and Martin Sauter — Karlsruhe Institute für Technologie

Many user analyses in High Energy Physics (HEP) have a huge demand for computing resources. This demand is unpredictable due to peak loads, which makes it challenging to use the resources efficiently.

Additional computing resources that are not dedicated to HEP usage, so-called opportunistic resources, help to cover the resource demand. We allocate these resources from HPC centers or cloud providers on demand and integrate them transparently into our institute batch system. This results in a flexible cloud-like system with a huge number of heterogeneous resources and a single point of entry.

To improve the utilization of these resources, we developed COBalD (Opportunistic Balancing Daemon), a framework that makes decisions using feedback loops. This allows for reacting dynamically on the current demand of resources while adjusting the amount of resources based on their utilization. For the integration of resources from different resource providers, we combined CoBalD with a multi-agent resource manager called TARDIS (Transparent Adaptive Resource Dynamic Integration System) in order to interact with various providers. TARDIS enables reacting on the current demand in a higher granularity which results in a higher resource utilization.

In this contribution, we give an overview of our developments as well as the experiences with our system.