Biogen has picked Blackstone Technology Group of Worcester, Mass., to construct a distributed compute farm system to accelerate its drug discovery efforts.
Rainer Fuchs, Biogen’s director of research informatics, said Biogen is expecting the new framework, which will provide 170 CPUs of compute power, to help it systematically analyze the complete human genome and run thousands of Blast and other searches in a timely manner.
“It’s a throughput issue – the farm gives us the ability to send significantly more data through the system than ever before,” said Fuchs. With so much publicly available genomic data, the challenge is to analyze the data in a reasonable timeframe, he added.
The Cambridge, Mass.-based Biogen also anticipates that the computing design will allow it to update its databases efficiently, a process that is done nightly as new data from the public databases are published.
“The solution will allow us to maintain an up-to-date view of the human genome that allows our scientists to identify the most promising targets,” said Fuchs.
Such a compute farm also has built-in resilience because if one node stops working, jobs can be routed to other processors in the farm, said Fuchs.
Blackstone is currently installing the new system, which is expected to be operational by December. The compute farm will be based on the Linux operating system and Intel-based processors and will interoperate with Biogen’s existing Unix infrastructure, which consists of Sun Microsystems and SGI hardware. Blackstone will also add some new Sun hardware to Biogen’s computing facility.
Financial terms of the deal were not disclosed.
Biogen intends to use Blackstone’s SmartBlast tool to optimize Blast searches in the new architecture as well as its SmartCache data management tool to speed data flow across the farm. The SmartWatch software will also be deployed to monitor CPU usage rates and efficiency levels.
Blackstone will also provide Platform Computing’s LSF load-balancing software to streamline processing across CPUs in the compute farm.
Fuchs said that Biogen chose to work with Blackstone because it has a combination of technical expertise and bioinformatics domain knowledge. He pointed to people like William Van Etten, principal bioinformaticist at Blackstone and a former Whitehead Institute researcher, who have experience installing and using compute farms.
Fuchs liked the way Blackstone handled Biogen in part because the consultancy was very open-minded about the technical solutions. “They didn’t try to push us towards one particular setup,” said Fuchs, who declined to disclose the names of the other compute farm providers Biogen evaluated.
Biogen didn’t have any strong concerns about the technical viability of the compute farm, especially since Celera Genomics is a Blackstone customer. “That made me feel more comfortable about it,” said Fuchs. Genome Therapeutics is also a Blackstone client.
Blackstone expects to see more companies investing in compute farms as more compute power is needed to manage and analyze the growing pool of genomic data – which, according to Van Etten, is doubling every six months. These volumes of data “are almost too large to be handled on a reasonably priced mainframe,” he said.
In Van Etten’s view, compute farms scale more easily, are more cost-effective, and use computational power more efficiently than mainframes can.
Compute farms can greatly outperform traditional network systems, some of which only enable companies to use 30 percent of the total available compute power, said Van Etten. He said Blackstone would be able to more than double that efficiency rate, up to rates of 70 to 90 percent by breaking tasks into pieces and assigning them to available processor resources.