ACADEMIA

Faced with scientific and engineering problems of increasing complexity and size, the Mississippi Center for Supercomputing Research (MCSR) has acquired a 64-processor SGI Altix system from Silicon Graphics to significantly "raise the bar" for computing resources available to more than 800 researchers throughout Mississippi. Serving students, faculty and researchers at eight Mississippi universities and collaborating with two Department of Defense facilities, MCSR acquired the new Altix system to tackle large-scale calculations, such as those common to computational chemistry and molecular modeling applications. In recent years, use of compute-intensive applications like Gaussian and NW Chem has grown more prevalent at MCSR, pushing the center's existing systems to their limit. Because MCSR users conduct both basic and applied research in such demanding disciplines as atmospheric modeling, high-energy physics, and cancer and AIDS research, the facility increasingly faced significant delays in processing large-scale jobs. Already home to several supercomputers-including vector systems and a 219-processor Beowulf cluster-MCSR sought the performance, scalability and ease of administration available only with a powerful shared-memory architecture based on Linux operating system and Intel Itanium 2 processors. MCSR's new Altix system includes 64GB of system memory and 2.4 terabytes of local FibreChannel disk space. MCSR installed the system in June and, after extensive benchmark tests, made it available to users at various universities, including the University of Mississippi, Mississippi State University, the University of Southern Mississippi, and Jackson State University. "Smaller programs run well on our cluster, but over time we found our users were facing large problems that simply were too much for even our most powerful systems," notes David Roach, MCSR director. "Rather than extend the same technology we already had in place, we decided we needed to raise the bar in terms of what we could accomplish." Having followed the SGI Altix family of servers and supercomputers since its introduction in January 2003, MCSR staff already knew that applications such as Gaussian, NWChem and AMBER performed exceptionally well on Altix. For instance, Dr. Randy M. Wadkins, assistant professor of chemistry and biochemistry at the University of Mississippi, uses AMBER 7 to calculate molecular dynamics trajectories for carboxylesterase enzymes that activate a number of important drugs. Preliminary Altix results show a nine-fold increase in the speed of these types of calculations as compared to Wadkins' own RLX Technologies cluster. The resulting performance boost should allow calculations that would have taken more than a week on Wadkins' 32-node cluster to finish in just over a day on MCSR's Altix system. "With the chemistry applications and some of the CFD codes developed by our researchers, we had jobs that required us to hold the entire data set in memory," Roach said. "Faculty were reporting that many of these jobs required more memory than the cluster could provide, and they simply wouldn't finish at all on our older systems. To ensure that we could address the needs of our user base for the foreseeable future, we really needed to push our environment to the next step, and we quickly concluded that Altix was the best solution for us." MCSR operates at one-quarter to one-tenth the budgets of similar facilities in other states while managing $43 million in funded research, so the price/performance advantages of Altix proved a key purchase factor. "We're funded by the state of Mississippi and carefully measured on the return we provide our users based on the technology investments we make," Roach added. "The fact that Altix is based on industry standard technologies like Intel Itanium 2 processors and 64-bit Linux has a positive effect on the affordability of the system, while the global-shared memory architecture delivers what we need in terms of sheer performance. All this combined to give us the most optimal value possible from a new system purchase." The expansion capability of Altix was another important factor for MCSR. "Our funding sometimes comes in small amounts, which means we need to be able to grow our resources incrementally," said Roach. "SGI keeps raising the bar on the number of processors that Altix can incorporate, and we hope to scale our new Altix even further in the future, whether in terms of CPUs, memory or disk space." MCSR's Altix system, known as "redwood" to users and administrators, operates all 64 processors within a single system image-one instance of the Linux operating system. This eases administration tasks for MCSR, which has occasionally faced challenges in administering more than 200 nodes within its current Linux Beowulf cluster. "There's no doubt that administration is easier on an SSI," Roach added. "In our environment and with our user base, there is a need for a large SSI just as there is a need for a cluster."