Linux Cluster: Upgrade of the parallel computing capacity

[LRZ aktuell]

LRZ intends to take two new systems into user operation in its Linux
 Cluster infrastructure within the next weeks; these new systems are
 focused on efficient execution of large scale parallel programs.
 
 The first system is an Infiniband-connected cluster of 16-way shared
 memory systems with more than 2500 cores delivered by MEGWare, and the
 second one consists of SGI Ultraviolet (UV) large-scale shared memory
 systems with more than 2000 cores in total.
 
 It is expected that the Infiniband cluster will become available for
 regular user operation first, followed by the UV system somewhat later.
 However, the existing UV will be retired from user operation soon
 because the space it presently occupies will be needed for the new
 system. Furthermore, the Infiniband cluster will need to be moved into
 the new LRZ infrastructure later in summer, so there will be a
 prolonged interruption of operation some weeks after its initial user
 operation.
 
 Users are advised that all new systems, which are targeted to running
 large parallel jobs, will use SLURM as a batch scheduler. Existing SGE
 scripts will not work on the new clusters - they will require rewriting
 of the control section, and will also need modification of the program
 startup procedure in the script section. Documentation for the new
 batch system will be available via the Linux Cluster page (/www.lrz.de/
 services/compute/linux-cluster) once user operation starts. The usage
 instructions for startup of MPI programs will also be updated.
 
 We hope that this capacity upgrade will reduce the rather long waiting
 times which nowadays are observed for parallel jobs. We intend to keep
 you informed about the status of these new systems via updates to this
 document.


 This information is also available on our web server
 http://www.lrz-muenchen.de/services/compute/aktuell/ali4065/

 Reinhold Bader