Speaker
Mr
Sergey Chechelnitskiy
(Simon Fraser University)
Description
SFU is responcible for running two different clusters - one is designed for WestGrid internal
jobs with its specific software and the other should run Atlas jobs only. In addition to
different software configuration the Atlas cluster should have a diffener networking
confirugation. We would also like to have a flexibility of running jobs on different
hardware. That is why it has been decided to run two clusters in virtualized environments.
Extensive tests of running CE and SE in a Xen-virtualized environment have been performed.
The following configuration has been selected for CE.
The performance of Xen virtual machines has been found to be excellent.
Each WN is running two virtual machines - one is running internal WestGrid jobs and the other
is running Atlas jobs. Moab is scheduling jobs for both clusters. We work closely with Cluster
Resources on different moab configurations. We plan to implement memory management when Moab
changes the amount of assigned memory between two virtualized WNs running on the same hardware.
We plan to create an automated system of Xen images replication from one piece of hardware to
another. An optimal Xen and Moab configuration for a geterogeneous clusters should be
developed.
Different tests of running dcache SE in Xen environment have been performed.
Primary author
Mr
Sergey Chechelnitskiy
(Simon Fraser University)
