Dell’s involvement in life sciences computing was relatively quiet until the company’s recent push to win the contract for Jeff Skolnick’s 4,120-processor cluster at the Buffalo Center for Excellence in Bioinformatics. To get the scoop on the computing giant’s view of the industry, GT’s Meredith Salisbury caught up with CEO Michael Dell and Reza Rooholamini, director of clustering for Dell’s enterprise systems group. Dell’s cluster group has been around some four years and its life science installations are used at the University of Alabama, Birmingham; Georgia Institute of Technology; University of Pennsylvania; Rutgers University; University of Tennessee; and the University of Texas, among others.
Why are bioinformatics customers using clusters instead of traditional big iron supercomputers?
DELL: Customers — whose budgets once couldn’t support upfront or ongoing maintenance costs of a supercomputer — can deploy high-performance computing clusters in their labs and on their campuses for approximately 10 percent of the cost of a traditional supercomputer. And the systems scale more efficiently as performance needs increase by adding additional low-cost, high-performance servers. Economics and technology such as these could very well eliminate the viability of the traditional supercomputer in three to five years.
Where will compute power and storage ability be three, five, 10 years down the road?
DELL: If you look at Moore’s law alone, we could theoretically have 128 times the processing power we have today in just 10 years. Another interesting way to evaluate how far these technologies have come is to look at the cost of a Gflops, or giga floating point operations per second, as the standard measure of scientific computing performance. Today we can deliver close to 800 Gflops for $1 million. Three years ago it was close to 400 Gflops for $1 million. Historically, this indicates that buying power will double every three years.
Bioinformaticists should fasten their seatbelts. Computers are expanding our research capabilities in ways that could dramatically benefit society, government, and commerce.
Another interesting trend to prepare for will be the evolution of these HPCC solutions into grids. As more and more individual researchers deploy their own clusters departmentally, we expect they will eventually link them into a grid. Of course, grid technology still needs to standardize more before this reaches fruition, but we are optimistic that this will happen in the next three to five years.
Where is the biggest challenge in IT for life sciences? From what other industries are the solutions coming?
DELL: Multiple challenges are emerging in the life sciences space, including managing larger and larger clusters, faster access to the large amounts of data, and ensuring that the interconnects keep pace with the sheer size of these clusters. This challenge will only increase over time as cluster sizes increase. Today, thousands of servers are being deployed in a single cluster; three to five years from now, we expect to see tens of thousands of nodes in a cluster working on a single problem or in a computing grid.
The solutions for addressing these challenges will come from technology companies working with academia, government, and industry. Industries such as financial services, manufacturing, and energy that are also deploying clusters to address data-intensive computing will need to share best practices to continue to move HPCC capabilities to the next level.
Academia has traditionally led in this endeavor. Many of the universities throughout the world are working together today to improve HPCC solutions with new technologies, such as Clemson University’s Parallel Virtual File System project and the University of California, Berkeley, with their Ganglia project for monitoring of clusters of all sizes.
Some people argue that clusters are a transition technology, bridging the gap until the next big supercomputer advance. Where do you see clusters fitting in?
ROOHOLAMINI: If you look at history, there’s an evolutionary path of supercomputing. In the ’60s and early ’70s we had proprietary supercomputers like Cray. In the mid-’80s and even the early ’90s we saw the next step, supercomputers that were built using microprocessor technology. In the late ’90s we see the next step: that is, supercomputers built using the cluster technology. That architecture allows for new technology to be injected into the system very quickly and easily and provides an environment where heterogeneous architecture could coexist.
We have customers who, because of the cost of owning a supercomputer, just were not able to run [life sciences problems]. But now with clusters they can afford to put in an eight-node or a 16-node cluster and basically have a supercomputer right in their office. We also have customers [who] were running their applications on proprietary supercomputers — they are gradually decommissioning and moving to clusters for bioinformatics. We look at bioinformatics as a key application domain.
What’s the promise of Linux? Will it keep going strong?
ROOHOLAMINI: We see people in both camps. We have customers who, for some reason or another, want to have a cluster with Windows. We also have customers who prefer Linux. We support the Linux community and believe it’s going to be around for a long time.