OnDemand (Rocks-131) Cluster User Guide
The OnDemand cluster is a Rocks cluster with Intel dual-socket, dual-core compute nodes. The 2.0 GHz, 32-way nodes have 8 GB of memory. OnDemand has a nominal theoretical peak performance of 2.4 TFlops. The cluster has 30 TB in its IBRIX parallel file system. Jobs are scheduled by the Sun Grid Engine.
Recommended Use Guidelines for OnDemand
Access to the OnDemand cluster will be granted to selected research projects with a well-defined need for immediate service for infrequent but time-critical, emergency-support applications with large-scale parallelism. Examples might include simulations that would need to run in response to natural phenomena, such as earthquakes or storm systems.
Requests for use of the OnDemand cluster in this mode should describe why runs are time-critical and cannot wait in standard batch queues or be scheduled in advance. Projects that adequately justify a plan to use a mix of time-critical and non-urgent runs with the same application code(s) may be granted time on OnDemand for both types of runs, if time is available.
Please contact Dongju Choi via consult@sdsc.edu, for discussing your groups needs.Access to the Rocks-131 Cluster for Jobs that Are Not Time Critical
The remaining CPU-hours on the cluster resource will be available to regular jobs through the batch queue until an on-demand job is submitted. Special access to the cluster for general, non-urgent use, including StarP jobs, is available to researchers at the University of California. Contact Subha Sivagnanam via consult@sdsc.edu, for access.
Job Priorities
Jobs that are time-critical are given priority over other jobs on this cluster. Note that any non-urgent runs on this system may be preempted for urgent, time-critical runs.
