SGE (new cluster)

Access to the new cluster is via the pool.

The new cluster (lxb682-...) is available via SGE only and runs under LXadmin.Debian Squeeze 64 bit only. It has no NFS mounts (no /u, /d, /s, /misc or whatever) only LXadmin.Lustre. Prefered working directory is /tmp .

The software is available via the Cern Virtual Machine File System cvmfs . A so called build server will be available for the experiments, so they can install their software by themself. Software from smaller departments or generall software will be deployed by the IT.

Unlike in the old LSF cluster we will not provide experiment specific or preemptive queues anymore, instead of this there are queues with different resource requirements/limits.

The priority giving for jobs/users/experiments will be done via FairShare, means every department will get an amount of shares and the jobs belonging to this department will get a priority depending on these shares and the resource usage of this department in the past.

Inside a department there can be projects or users defined geting a defined sub amount of the department shares. If nothing is defined inside the department all jobs belonging to it will be sheduled in a first come first serve order.

For answers on how to use SGE have a look at :

-- CarstenPreuss - 24 Oct 2011
Topic revision: r2 - 2011-11-07, BastianNeuburger