By Meredith W. Salisbury
Don’t blame Kevin Howard for trying to ruin your weekend at the beach. If he gets his way, soon you’ll be spending your between-swim time running Blast queries and other bioinformatics applications on your PDA.
Howard, a founder of Massively Parallel Technologies, has spent nearly six years putting together what he calls “a new class of high-performance computer” that dramatically lessens the time to run queries for programs like Blast and which has just become accessible to scientists through a Web interface.
When Howard started looking at the high-performance computing industry, he says, “I came to the conclusion that something must have been radically wrong. If the vendors were correct that you needed to have a very fast point-to-point connection and very low latency, then the human brain just couldn’t function.” Backed by grants from the US Defense Advanced Research Projects Agency, Howard and his startup team began to study the brain’s communication abilities, using that information to find ways to improve communication speed in machines. What he found — a way to increase parallel data movement within the system — means that a computer can run using inexpensive switches and other communication hardware, drastically lowering the cost of computing.
Howard’s company built up the infrastructure in Louisville, Colo., in what it calls a Virtual Power Center. The business model, providing “on-demand” high-performance computing, grants customers access to the compute resources for the price an average person might pay per month for cell phone service, Howard says.
Customers don’t need any equipment beyond what’s already on their desktop, he adds. Data is stored and processed on Massively Parallel’s infrastructure, and users can view the results of their queries immediately, rather than waiting for the data to be sent back over the Internet. Right now, anyone with a Web server can use the technology; the Massively Parallel team is working on making the tool accessible via PDA as well. “What I hope to have,” Howard says, “is supercomputing at the beach.”
Massively Parallel Technologies has started its offering with an implementation of NCBI’s Blast algorithm, and plans to add others — such as Fasta, HMMer, and Smith-Waterman — later this year. Howard says the boon to users, aside from a significant speedup in getting answers, will be “a seamless tying together of all of these applications.” He says users will be able to run their Blast queries, select the results of interest, and then move those directly into a molecular dynamics program, for instance — without having to translate any of the data. Users see up front “how long it’s going to take and how much it will cost, and you can choose to run it there or not,” Howard says. “We really want to be the choice in doing computational biology because you … won’t be able to beat the price even with your own equipment.”
The Virtual Power Center is just taking beta testers now, and the company plans to spend as much time as necessary in beta mode to make sure everything’s in place. “Since we took our time learning how to [build this],” Howard says, “we’re also taking our time to make sure this is what the users really want.” One early beta tester, Jan Jensen, an associate professor at the University of Colorado Health Sciences Center, says, “We were able to conduct preliminary work for a grant request in two and a half weeks — an effort that commonly takes two years to complete.” Jensen credits the company’s implementation of Blast as “a key contributor” to the speedup.
Howard says his company is still looking for beta testers, and he encourages anyone to come try out the service. People who are interested should e-mail him at [email protected] and give a basic indication of the databases they’d like to use and what kind of work they’re doing. “We e-mail them an access code and they can have at it,” Howard says.