886 A Sustainable High Performance Computing System for Teaching and Research

Thursday, 14 January 2016
Steven Boring, San Jose State University, San Jose, CA; and S. Chiao and A. Bridger

A sustainable high performance computing (SHPC) cluster was built using computers that were being replaced from a computer lab. Computer servers are typically upgraded or replaced every few years. However, those machines still have fast processors and life to run in a cluster for a few more years. Thus extending the life of computers by several more years. The reuse of these computers well cut down on the cost of purchasing new computational servers or HPC systems. Since there is always a steady supply of used computers, the cluster can be continually expanded and upgraded as needed. Also the use of open source cluster software (e.g., Rocks cluster software) will keep the costs low. With the use of scheduling software, multiple users may use the cluster more efficiently. The cluster has been built using 9 computers, 8 used for compute nodes and 1 as the frontend node. All the Computers have dual core processors and 4GB, resulting in a cluster that has a total of 18 cores and 36 GB of RAM. A similarly configured server could cost more than $4000 so there is a great savings using the older systems. The computing power of this system can easily be increased by plugging extra computers into the cluster. The power used to run the cluster is generated via a co-generation plant on campus to further reduce the cost.

A benchmark modeling project being done of the SHPC is a study of hurricanes in the Caribbean. The Meteorological conditions of 5 hurricanes (Charley, Dean, Denis Emily and Ivan) passing through the southern Caribbean are simulated with WRFV3.7 using two different reanalysis data sets (GFS and ECMWF) at high resolution (5km) and the outputs compared with each other.

In addition to support regular modeling tasks, this SHPC system can be incorporated in to the curriculum in the area of parallel coding, MPI (message passing interface), as well as the HPC architectures. Students will be able to gain basic knowledge on how to work with computational cluster and supercomputers.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner