Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

HK: Fachverband Hadronen und Kerne

HK 9: Accelerators and Instrumentation I

HK 9.5: Vortrag

Montag, 16. März 2009, 15:15–15:30, H-ZO 80

Exploiting Virtualisation at Alice HLT — •Stefan Boettger, Volker Lindenstruth, and Udo Kebschull — Kirchhoff-Institut für Physik, Heidelberg

The Alice HLT cluster is a computing farm intended to do on-line event processing for the ALICE Experiment at CERN. It is known that at run-time of the experiment there are phases where few or no data is available for processing. The same applies for maintenance cycles of both the experiment and the cluster. With respect to the costs of maintaining and running such a cluster there is the need to maximize the usage of this computing facility. Therefor the usage of those idle times with third-party off-line physics computations was proposed. To satisfy the constraints of on-line and off-line applications and to avoid interferences, the usage of os virtualisation has been evaluated. Moreover, a comparison between Vmware-Server and Xen concerning system-level performance and usability was done. Results show Xen to be superior to Vmware-Server regarding general performance measurements, yet being more complicated to install and maintain. It could be shown that os virtualisation is a feasible way of using idle cycles, avoiding application interferences and maximizing cluster usage by suspending and migrating of virtual systems. A future extension based on automated switching between on-line and off-line data processing using the SysMES cluster management framework is proposed.