Hamburg 2016 – scientific programme

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

T: Fachverband Teilchenphysik

T 97: Grid-Computing

T 97.5: Talk

Thursday, March 3, 2016, 17:45–18:00, VMP8 SR 05

Dynamic integration of remote cloud resources into local computing clusters — •Georg Fleig, Günther Erli, Manuel Giffels, Thomas Hauth, Günter Quast, and Matthias Schnepf — Institut für Experimentelle Kernphysik, Karlsruher Institut für Technologie

In modern high-energy physics (HEP) experiments enormous amounts of data are analyzed and simulated. Traditionally dedicated HEP computing centers are built or extended to meet this steadily increasing demand for computing resources. Nowadays it is more reasonable and more flexible to utilize computing power at remote data centers providing regular cloud services to users as they can be operated in a more efficient manner. This approach uses virtualization and allows the HEP community to run virtual machines containing a dedicated operating system and transparent access to the required software stack on almost any cloud site.

The dynamic management of virtual machines depending on the demand for computing power is essential for cost efficient operation and sharing of resources with other communities. For this purpose the EKP developed the on-demand cloud manager ROCED for dynamic instantiation and integration of virtualized worker nodes into the institute’s computing cluster. This contribution will report on the concept of our cloud manager and the implementation utilizing a remote OpenStack cloud site and a shared HPC center (bwForCluster located in Freiburg).