University of Notre Dame

Story Details

High-performance computing (HPC) has become the engine of new discoveries at universities and other leading research centers worldwide. Keeping the life-cycle costs of HPC systems under control, even as the demand for computing power soars, has become a major challenge for the information technology experts who support the research community. By reducing the time and costs required to design, purchase, implement, and expand HPC systems, research computing centers can maintain their primary focus on developing new applications and supporting users.

Toward those goals, the University of Notre Dame in 2005 began planning an ambitious expansion of the HPC capacity at its Center for Research Computing (CRC). The CRC supports research in fields as diverse as biomedical research, molecular modeling, computational fluid dynamics, economic analyses and computing itself.

Additionally, user demands for more computing power were increasing both from within the Notre Dame research community and from the pending participation of the CRC in the Northwest Indiana Computational Grid. Funded by the U.S. Department of Energy, the grid initiative will pool the compute cycles and storage capacity of Notre Dame, Purdue University and Argonne National Laboratories into a single resource that can be tapped by researchers throughout the region.

After evaluating several technology options, including solutions from HP, Notre Dame selected an HPC server cluster from Sun Microsystems. Sun V65z and V20z servers had already proven themselves for reliability and cost-effective performance in two HPC clusters at Notre Dame totaling 144 compute nodes. Now Notre Dame turned to Sun for computing power on a much larger scale, and for ready-to-deploy servers that needed to be delivered and brought online within a tight time frame.

The Sun solution, based on Sun Fire X2100 and Sun Fire X4100 servers, gave Notre Dame more teraflops of computing power for the dollar, as well as a ready-to-deploy architecture that dramatically simplified and accelerated the process of adding new capacity. The Sun architecture also offered an excellent ratio of performance to power consumption. In addition, the Sun servers met Notre Dame's requirement for servers based on AMD Opteron processors—a key to maximum price-performance.

The cluster now consists of 360 nodes. This includes ten racks of 360 Sun Fire X2100 servers and one rack of four Sun Fire X4100 servers used for command and control. Based on Sun Grid architecture, the solution uses Sun N1 Grid Engine software to monitor and select the optimal usage of computing resources for batch processing. To keep the entire infrastructure running smoothly, a Sun Hardware Service Plan provides Notre Dame with the added assurance of online and onsite support.

Sun Customer Ready Systems (CRS) team configured the solution for Notre Dame and met their requirements for rapid deployment. These integrated systems (or compute pods as they have been aptly named) left the factory ready to deploy, speeding implementation times and reducing ongoing maintenance and upgrade costs. In one phase of the deployment, Notre Dame was able to order two new pods consisting of 72 new Sun Fire X2100 server nodes, add them to the existing cluster, and begin using them to support research within two weeks.

By contrast, Notre Dame estimates that it would need to dedicate several employees weeks of work in order to architect, configure, install and tune a comparable HPC cluster using technology not already integrated and proven effective by Sun. The factory-integrated Sun systems free Notre Dame's HPC staff to focus on applications and users while reducing TCO through lower maintenance, support and upgrade costs. The small footprint and low power requirements of the Sun servers are also helping Notre Dame manage data center costs over the long range.

Thanks to the Sun server cluster systems built in the Sun CRS factory, the CRC at Notre Dame has expanded its capacity by 250% within its desired time frame and budget. Notre Dame can now repeat its order anytime and add new capacity of ready-to-deploy compute pods in a matter of weeks. Like the Sun Grid architecture, research computing at Notre Dame now has the potential for limitless expansion.