Challenge
Build a high-performance computing cluster capable of simulating large-scale flu pandemics in complex, real-world populations.

How Silicon Mechanics Helps
Delivering cost-effective, best-fit server configurations that help maximize total computing performance per dollar spent.

Benefits
Researchers have the computing power they need to model massive flu pandemics — and to determine the best ways to mitigate and contain them.

Getting the Drop on Avian Flu

When an elite team of researchers needed a high-performance computing cluster for combating one of today's most critical public health threats, they turned to Silicon Mechanics to help them maximize the value of their HPC budget and speed their time to results.

The world may be on the brink of a major flu pandemic that would sweep the globe with unprecedented speed on the wings of today's international air flights. For now, "bird flu" is not transmitted easily from person to person, but that could change quickly, and communities and healthcare organizations around the world need to be prepared.

Shufu Xu
Systems Analyst
Fred Hutchinson Cancer Research Center

Saving Lives through Fundamental Research

An effective response to this threat requires a deeper understanding of how flu viruses spread in complex societies. To provide that understanding, biostatistician Dr. Betz Halloran, systems analyst Shufu Xu, and their colleagues at the Fred Hutchinson Cancer Research Center have created software that can accurately model the process.

According to Dr. Halloran, "Once an outbreak occurs, there are many ways to contain it and limit its impact, and they all have different costs and practical limitations. Our research focuses on understanding the effects of targeted interventions, in order to determine the best strategies in particular scenarios." The results are helping public health officials prepare more effectively, so they can make the best use of available resources — and possibly save millions of lives.

Maximizing Performance per Dollar

In the research team's largest simulations, every man, woman, and child in the US is represented. That includes more than 280 million people, and it takes tremendous computing power to accurately model their movements and interactions. Until recently the research team borrowed time on other organizations' supercomputing systems, but they needed a dedicated in-house system to speed their research.

Xu has extensive expertise in high-performance computing and had a clear vision of the design requirements. After interviewing a number of major server manufacturers and system integrators, a colleague recommended he speak with someone from Silicon Mechanics. It was a good tip. According to Xu, "Our models can take up to 120 hours to run, so we need all the performance we can get. Silicon Mechanics gave us the configurations we needed at the best price."

"That's not unusual," says Art Mann, Vice President of Business Development for Silicon Mechanics. "Our goal is to deliver the precise configurations each customer needs, without any unnecessary frills or over-configured components that raise the price." Whether they're deploying a single server or a cluster, that approach delivers better value for customers. "The lower cost was important to us," says Dr. Halloran. "It allowed us to build a larger cluster within our budget, so every simulation runs faster."

Dr. Betz Halloran
Biostatistician
Fred Hutchinson Cancer Research Center

A Powerful Cluster and Responsive Service

When first deployed, the system included 23 servers, each configured with two dual-core processors. Since then the cluster has been expanded twice, and now includes 90 servers (360 CPUs), interconnected by a stack of six Gigabit Ethernet switches. The "head node" for the cluster has 1.7 terabytes of storage, to handle the massive amounts of data collected in the simulations.

The high-performance cluster is a critical asset for the research team, so keeping it up and running at all times — and at full power — is important. As Xu notes, "With any large s erver cluster, there will always be occasional hardware failures, so service was another factor in our purchasing decision. Silicon Mechanics has been very responsive. Whenever we have a problem, they get replacements to us within a day — two at the most. With the other vendors we considered, the turnaround would have been at least three days."

Art Mann
Vice President of Business Development
Silicon Mechanics

The Work Goes On — Only Faster

For now, Halloran and Xu have all the computing power they need and are focused on conducting their research, extending their models, and refining their software for even faster performance. They're pleased with their cluster and with the support they've received from Silicon Mechanics. "We get a lot of inquiries from colleagues who are interested in building high-performance clusters," says Dr. Halloran. "We always suggest they include Silicon Mechanics in their short list of potential vendors."