The challenge

The National University of Singapore (NUS) is a top research university that has an insatiable appetite for computing power to push the boundaries of computation-based research. NUS has thousands of computers across its campus for teaching and learning, and these computers have a lot of unutilised capacity. These computers are as powerful as the supercomputers 20 years ago. The demand for additional compute power from research laboratories and departments has been increasing, but the cost of equipping each for power-hungry applications was prohibitive.

The solution

NUS decided to harness the unutilised capacity from existing servers and PCs in the campus to set up a supercomputing grid. NUS and SCS signed a memorandum of understanding to collaborate on this project.

SCS' involvement in the project includes infrastructure deployment, consisting of system software for the servers and agent software for the node computers; hardware deployment for the servers managing and controlling TCG@NUS; and development and enabling of applications to be deployed on TCG@NUS.

The grid management infrastructure was implemented using United Devices'Grid MP Platform. The Grid MP platform creates a virtual infrastructure that enables existing computer systems and networks to be translated into shared resources.

Originally aggregating 800 computers contributed through voluntary participations from more than ten departments in NUS, TCG@NUS is targeted to extend to 3,000 nodes by 2007 to produce a theoretical peak performance of 13 Teraflops.

The benefits

TCG@NUS is Singapore's first large-scale campus-wide grid implementation which harnesses idle computing capacity to generate high-speed computing power. TCG@NUS is a groundbreaking regional project, as it is the largest of its kind among educational institutions outside of the US and Europe.

TCG@NUS has catapulted NUS into the ranks of those research institutions capable of carrying out massive computations. The creation of this unprecedented compute power opens up new areas of discovery and high impact research in time frames previously unimaginable. Availability of large-scale capacity allows scientists to widen their scope and deepen their focus, leading to better quality of research results.

The investment of about S$300,000 creates total capacity of between 6 TFLOPS and 12 TFLOPS. S$3-4m would be required if a traditional hardware-centric solution was used for the same capacity. This translates into a very good return on investment if grid computing is used.

In aero-genomic research, TCG@NUS was able to align 22,000 genome sequences against 3,461,799 sequences (15GB of data), in two days. On a single machine, this computation cycle would have taken several weeks to complete.

NUS has also been able to undertake a cancer research project that involves screening 80,000 chemical compounds, thanks to TCG@NUS. The AutoDock project, a docking application for drug discovery, consumed 165,000 CPU-hours in a month, using TCG@NUS. If the same work had been processed in NUS' 64 node cluster, it would have taken four months of non-stop compute time. And if this job had been run on a PC, it would have taken 18 years to complete!