Who Needs Supercomputers?
Each workday, the scientists at Monsanto Co. (MON ) try to develop new ways to genetically alter food staples such as rice, corn, and wheat to make them more resistant to the harsh elements of drought and cold. That kind of DNA tinkering requires the painstaking matching of genetic patterns to other known sequences so scientists can figure out what traits yield better crops. The amount of data collected by Monsanto scientists on animal and plant genes fills CDs that would stack 500 feet high--enough to send even the most powerful supercomputer into cardiac arrest.
That used to be a drag for researchers because it meant they could analyze only 10 to 50 genes per year. "With our big computers, gene-analysis jobs were taking up to six weeks to finish," says Mark Trusheim, president of Cereon Genomics LLC, a Monsanto subsidiary specializing in genetic research. "That was a major bottleneck."
Two years ago, Monsanto made a breakthrough, though it wasn't genetic. The company is pioneering a technology called grid computing, software that lets users plug into computing power on the Internet or private networks as easily as electricity can be drawn from the power grid. Instead of spending tens of millions of dollars buying bigger computers, Monsanto taps hundreds of smaller--and less costly--Compaq Computer and Sun Microsystems machines already in use at its St. Louis headquarters and Cereon's offices in Cambridge, Mass. Grid software tackles a problem by slicing it into thousands of tiny pieces that can be solved independently and then spliced back together. When Monsanto's machines aren't handling other tasks, they get a small calculation to work on--lickety-split. Now, a gene-analysis job takes less than a day, and Monsanto can examine thousands of genes a year--a fiftyfold increase over what was possible five years ago.
The rest of the computing world may be on the edge of a similar pop in power. Grid technology, born in the halls of academia and government research labs, is going mainstream. On May 23 at the Electronic Entertainment Expo in Los Angeles, startup online game producer Butterfly.net Inc. and IBM (IBM ) introduced the first computing grid for the Internet game market. With a grid setup, Butterfly can handle an unlimited number of online players by automatically calling on other machines at an IBM data center when usage spikes and comes close to overwhelming the service.
Grid computing is gaining popularity in other corners of Corporate America as well. Carmakers General Motors Corp. (GM ) and Ford Motor Co. (F ) analyze vehicle crash tests using a grid network. Pharmaceutical giants Adventis (AVE ), GlaxoSmithKline (GSK ), and Pfizer (PFE ) evaluate human disease and develop treatments. And aircraft-engine maker Pratt & Whitney tests computer-aided simulations of engines, slashing in half the time and cost it once took to conduct these tests. "We see grids as the next big thing in computing and the evolution of the Internet," says IBM Vice-President Irving Wladawsky-Berger.
Indeed, grids could one day transform the economics of computing. The most complex computer jobs typically require a supercomputer, which can cost $30 million for a middle-of-the-road machine. If a company doesn't own one of these digital Ferraris, the task can take an excruciatingly long time. But by aggregating the power of smaller, more affordable computers, grids can outmuscle the largest supercomputers for a fraction of the price--around $25,000 for the basic software and services. "Many companies are trying to pool computer resources," says Rick Hayes-Roth, chief technical officer for Hewlett-Packard Co.'s software group. "That saves money. That's what companies are looking for these days."
There are plenty of pesky challenges for grid computing to overcome--from managing computer resources to prevent gridlock to making sure security is tight. The fact is, piping computing tasks across a network increases the opportunities for hacker intrusions, especially if the network being used is the Internet. And with a large network of computers working on a common problem, making sure they have the right data can be tricky.
Companies must be willing to tinker to get their grids working smoothly. Incyte Genomics Inc., a Palo Alto (Calif.) biotechnology company, uses a 1,000-machine grid to do genetics research to build databases it sells to pharmaceutical companies. Originally, not all the machines on its grid were being sent information at the same time, which increased the chance for errors. Incyte fixed this by transmitting data over the network once so that all of the computers could pick it up simultaneously. "Keeping the data accurate is an issue," says Stuart E. Jackson, Incyte's director of bioinformatics.
When more of the kinks are worked out, grids have the potential to fit hand-in-glove with a host of new Web services. For example, a grid could be the backbone to orchestrate all sorts of information coming from different sources to an online travel site--everything from airline ticketing and hotel reservations to flight-delay notifications and traffic-congestion alerts. Such services can be provided today, only it requires loads of expensive hardware and software. A grid could turbocharge such a service by breaking it into simple tasks that could be tackled instantly by machines already on the Net. "Integrating Web services with grid computing will push computing toward being a utility that you can subscribe to on demand," says Ian Foster, a senior scientist at Argonne National Laboratory and a member of the Globus Project, a group that's developing technical standards for grids.
That day is getting closer. Early this year, IBM and the Globus Project made available a set of specifications to let businesses share computers and Web services over a grid. IBM says all its products will come with grid software as a standard feature. Rivals also are staking out their ground. On Apr. 9, HP unveiled software to set up grids within a corporate data center. And Sun Microsystems Inc. (SUNW ) offers grid software for free on its Web site. "We've been pushing the idea that the network is the computer," says Wolfgang Gentzsch, Sun's director of grid computing. "Now, the grid is the computer."
That's the bet Butterfly.net is making. The Los Angeles company has engineered a breakthrough in the online game industry. The way online games work today, customers are relegated to separate servers that support up to 4,000 users. That limits the number of players who can interact with one another and creates reliability and support nightmares. When a server is down or software needs to be upgraded, play comes to a dead stop. What's more, owning all the hardware necessary to carry a variety of games online is expensive.
So two-year-old Butterfly cut a deal with IBM to provide the horsepower for the online-gaming service. By running its grid software at an IBM data center with hundreds of servers, Butterfly can draw the computer power it needs when the number of users spikes. The software does this by automatically routing computer resources that are idle at the data center to help with the workload from the most popular games. For the online-game industry, "that's the Holy Grail," says Butterfly CEO David Levine. "This technology is going to be extremely important in the next decade."
Once security and other technical challenges are resolved, there may be no holding back grids. Some day, companies might be able to tap into vast networks of computers they don't even own--and pay a fee for what they use. Just plug into the grid and compute.
By Darnell Little in Chicago and Ira Sager in New York