Our cluster work is increasingly more active and complex. Research is the primary use for our clusters (see our projects), but as research in parallel statistical algorithms advances a new curriculum for statistics students emerges and increasingly education becomes an important component of our cluster effort. We provide support pages as a repository of answers to frequently asked questions and reference materials for our cluster users.
To provide for the widening range of needs we have several clusters for different purposes. Our dedicated cluster is where researchers run debugged and possibly optimized code. It is our fastest cluster. Scheduling is supervised and access is by permission only. Our sandbox cluster is where debugging, optimization, and learning can be done. It can also be used to run jobs but not as fast as on the dedicated system. This cluster is supervised and accessable by permission only. Our shared cluster is available to any affiliated department member. Only an account on the Statistics network is neccessary. On it one will find the same tools as on the dedicated and sandbox systems but access is not supervised and hence jobs compete with each other. To schedule time or obtain an account to work on our clusters write to cluster@stat.ucla.edu. You must be affiliated with our department in order to obtain access.
Many things are new to our cluster work. Besides the statistical application our work with the Mac OS X architecture is contributing new knowledge to the literature of cluster architectures and deployments. In the course of our work we are gaining much experience building Beowulf-style clusters with Mac OS X and the PowerPC architecture. This experience is described in the deployment section and in our support pages. (You will also find that some of our work is documented with the Academic Technology Services.) Specifications of the various clusters currently in operation in our department are described in the clusters section.
Work in progress.
Our deployment notes explain how to turn Mac OS X systems into Beowulf clusters.