Even as more and more bioinformaticists fondly refer to their homemade clusters as “supercomputers,” computer giant Cray is on a mission to bring real supercomputers into the market. President and CEO James Rottsolk recently identified the genomics market as a new target for his company. Cray has already tasted success here with its discovery that technology built into its supercomputers works like a charm on complex bioinformatics problems. GT’s Meredith Salisbury chatted with Rottsolk to find out about the lawyer-turned-computer guru’s plans to make a mark in what many would call a saturated field.
What are the compute problems in bioinformatics that most need to be addressed?
ROTTSOLK: The most scientifically challenging and economically important problems have yet to be addressed, in large part because this has not been feasible with computing capabilities available to date. Complex problems that have been addressed typically have not been solved exhaustively, with 100 percent validation.
In general we’re talking about problems regarding complex pattern-matching with very large databases. Some of the toughest problems we’ve heard researchers talk about are in the proteomics area. When we say our processor runs two orders of magnitude faster than an Alpha on this type of problem, that means not having to rely on a statistical analysis anymore.
What is it about the bioinformatics market that lets it keep going strong while other computer markets are slowing down?
ROTTSOLK: Bioinformatics is a relatively young, immature market with a lot of excitement and funding behind it. It is going strong because of the enormous value associated with its potential, although there has not been enough time to realize much of that potential yet. The real strength of this market will emerge as efficacious new drugs and therapies start coming out of the pipeline, disease mechanisms are clarified, and so on.
Much of the focus in bioinformatics has been on parallel systems — primarily clusters, which can be assembled for a fraction of the cost of an actual supercomputer. How will Cray compete?
ROTTSOLK: No one type of high-performance computer is best at everything. Clusters are great at solving large numbers of small problems, or big problems that lack complexity. There’s definitely a role for us. Unlike the others who are using RISC processors, we’re using processors that have this high computation rate. We also build systems that have very high memory and very high-bandwidth interconnect.
Clearly, not everyone has the financial wherewithal to buy a multimillion-dollar machine. You’ll see us offering several different alternatives. One would be to provide smaller machines with price points less than a million dollars. Another possibility is some kind of timesharing program. We could also enter into collaborative developments with people where we provide some of the resources and capabilities to use the machine. We’d like to find ways to make supercomputational resources available, and that’s going to require different approaches.
It’s a business where you tend to get what you pay for. But there’s no reason that people shouldn’t be able to obtain access to this resource.
Some people have said that clusters are a stop-gap measure. Where do you see them going?
ROTTSOLK: We see clusters as having long-term value for solving less complex problems. There are problems that cluster systems address fairly well, and in that situation, they may be preferred from a price/performance view. I’d certainly like to see them play a lesser role.
People will definitely be doing both [clusters and supercomputers]. No one is going to have a monopoly on this business.