Return to

Rice Cluster

Rice Group


In 2009, Xin-Liang Xu, then collaborating with Profs. Aaron Dinner and Stuart Rice, worked with JFI Computing to setup a new computational cluster. This cluster is very oriented toward highly parallelized jobs.

The cluster is divided into two queues. The nodes in the first queue are GPU nodes with nVidia Tesla cards and the Cuda development environment. The nodes in the other group, which were added later on, have four 12-core AMD Magny-Coeurs CPU's per node, for a total core count of 48 cores per node for the nodes in that group. In the latter group, a job can execute across up to 48 cores on the same node and share memory locally across that job on the same machine. This can make for some very fast parallel IPC.

The cluster was named "bajie", after Zhu Bajie, the famous half-pig, half-man character from the 1592 Chinese novel, Journey to the West (pictured). There was once a second login node named wukong, from Sun Wukong in the same story, but that machine never made it to the developed version of the cluster.

Zhu Bajie

This unusual configuration in which the entire cluster is aimed at various sorts of parallelization strategies has been helpful for jobs that are coded to take advantage of it, but new users must be aware that we have better computing resources for jobs that need fast serial performance. Both node groups on bajie underperform typical clusters on conventional serial execution jobs, since the main CPU's on the Tesla nodes are merely desktop i7's, and the AMD Magny-Coeurs CPU's are fastest only when taking advantage of having 48 processors all operating on the same big local memory map. If you need fast serial performance, contact JFI Computing about our other clusters.

Bajie has also been helpful with various legacy Fortran simulations inherited from the Rice Group's earlier cluster that they ran years ago at the Research Institutes building.

If you would like to develop simulations on bajie, please either contact JFI Computing, or the Rice Group for permission to use the cluster.