Providing the Speed and Power for Hypersonic and Molecular ResearchThe University of Queensland, Australia, hosts two strategic scientific hubs: the Centre for Computational Molecular Science (CCMS), which focuses on computational chemistry; and the Centre for Hypersonics, which specializes in aerodynamics, studying air flow at velocities of at least five times the speed of sound. Driving research at both facilities is a two-rack, 66-node Sun Customer Ready HPC Cluster. Customer Challenges
SolutionThe CCMS and Centre for Hypersonics augmented a 128-node grid cluster system with a two-rack, 66-node Sun Customer Ready HPC Cluster using Sun Fire V20z servers powered by the AMD Opteron processor Model 250. Business Results
Story DetailsThe University of Queensland’s CCMS works to develop critical capacity in molecular scale computations in interdisciplinary contexts, linking research in chemistry, biology, and engineering. Its work overlaps with that of the university’s Centre for Hypersonics in the area of chemical reactions that occur during travel at hypersonic speed. Both research units needed to augment their 128-node grid system. “We decided that the large-scale, shared-memory architectures were not really what we needed,” says Sean Smith, Professor and Director of CCMS. The old system was based on 32-bit Xeon processors and could not match the speed of the Sun Fire V20z powered by the 64-bit AMD Opteron processor Model 250. With its memory capacity up to 16 gigabytes, the Sun Fire V20z has changed the standard for high-performance computing. “The power of 64-bit processors has developed to the stage that on one Sun dual-processor node we have enough memory and disk space — and a fast enough processor — to handle about 90% of our calculations,” explains Smith. “We need to be able to run lots and lots of these calculations, and that’s why we decided to go with the cluster format with the Sun Fire V20z.”
"
Sun Customer Ready HPC Cluster gives you more compute nodes for less money than other machines. Sun delivers more mega floating-point operations per second (MFOPS) per dollar.
"
— Dr. Peter Jacobs, Senior Lecturer in Engineering, University of Queensland
The 66-node Sun Customer Ready HPC Cluster increased throughput by 20%, while taking up about half the physical space. Populated with 66 dual-processor servers, the system is generally accessed by 30 to 40 users at any given time. “Previously, we could only run a very few sample test conditions because it just took so long to run the computer,” says Richard Morgan, Professor and Director of the Centre for Hypersonics. “With this powerful Sun facility, we can now run a computer simulation in tandem with the experiment. Before, we’d have to do limited simulations in restricted areas of research. Now there’s no limit to what we can do; we’ve greatly expanded the regions of the flow that we can properly analyze and this has added real value to our experiments.” Dr. Peter Jacobs, a Senior Lecturer in Engineering who works at the Centre for Hypersonics and also interacts with researchers in computational chemistry, found that the new system works better. “Just the fact that the engineering and the architecture of these systems are more mature means that the system runs more smoothly than it did with the old grid.” In addition, the architecture has helped CCMS gain the ability to monitor the system remotely. “To be able to monitor what the running temperature is, for example, seems to be better with the new Sun Customer Ready HPC Cluster than with the old grid cluster.” “People using the system for design work have access to a hundred times more compute power than they had previously,” adds Dr. Jacobs. “The same is true for computational chemistry. There were experiments we wouldn’t have previously attempted because processing the data would take up to three years of compute time. Now we can conduct those experiments with perhaps two weeks of compute time.” |
Interested in Sun's Open Storage?
Download this paper today to learn about the tools, trends and key features of Sun's Open Storage solutions.
| ||||||||||||