Getting A Grip On Grid Computing

More companies are farming out jobs to groups of networked computers

Deep in the bowels of the USTA National Tennis Center in Flushing, N.Y., a dozen producers were busy updating the US Open Web site one September evening. They crowded around a TV and gasped as tennis star/fashionista Serena Williams strode onto the court wearing a studded outfit and knee-high black boots. Within moments, photos of her were up on the Web site. But while her outfit caused a stir, the technology powering the site was just as remarkable.

To handle surges of demand from sports fans, was plugged into a powerful computing grid run by IBM (IBM ). Big Blue had linked dozens of servers at its data centers so they behaved like one big computer. Any computer on the grid could serve up a US Open Web page one minute and do bank credit analysis the next. Customers get computing power when they need it. "It's truly an on-demand business," says Ezra Kucharz, the USTA's managing director for advanced media. "We get the resources we need. If we don't need them, others use them."

After a lot of hype and a long buildup, grid computing is going mainstream. Grids allow processing jobs to be split up and farmed out over a network to many computers so the work can be done fast on any machine that's available -- and organizations can trim their hardware and labor costs. For years most grids were used for scientific research. But that's changing. Analysts say 300 to 500 grids have been set up by -- or for -- businesses in the past year or so. And more are on the way. Of 149 large North American companies surveyed recently by Forrester Research Inc. (FORR ), 37% have set up grids, and 30% are actively considering it.


It's too early for any one or two tech companies to dominate the market, but they're scrambling for advantage. IBM, which has been selling supercomputer grid services for more than a year, announced on Oct. 1 a wide offering of utility-like computing services, including grids, from 11 IBM data centers around the world. Sun Microsystems (SUNW ) on Sept. 21 unveiled its own grid-on-demand service, with the wrinkle that customers can buy dollops of computing power for an introductory price of $1 per processor per hour, and even pay with a credit card.

Now the popularity of grids could get another boost. BusinessWeek has learned that in October a new industry group, the Globus Consortium, will be announced. It's rallying around a grid technology standard called Globus. Tech heavyweights including Hewlett-Packard (HPQ ), Intel (INTC ), Sun, and IBM are expected to be among the founding members. They plan to include the Globus Toolkit, open-source software used for building grids, as a common ingredient in their grid products. The goal: to link seamlessly hardware and software made by different companies. This move pleases tech buyers. "We want to put things together in a grid and not worry about lack of interaction," says Mark Cates, chief technology officer for investment banking at Wachovia Bank (WB ), which has a grid for pricing derivatives.

In fact, conflicting standards remain the biggest obstacle for grids. Many major tech companies have released grid products based on their own proprietary technologies. While most of them plan to adopt standards as they emerge, there's one major holdout: Microsoft Corp. (MSFT ). So far it's sitting on the sidelines. Its so-called Dynamic Systems Initiative, which is supposed to show up in products next year, presumes that people who want to tap into shared computers will use just one kind -- machines running Windows. "Our approach is, first and foremost, we'll make it work on Windows," says Charles Fitzgerald, general manager of platform strategy at Microsoft. "There's way too much standardization that takes place prematurely before they have done the engineering work." Most grid software allows grids to include Windows servers, but the process of setting up and operating those grids would be easier if Microsoft played along.

Corporations aren't waiting for the industry to achieve mind meld before building grids, though. DreamWorks SKG has a system that can hand off the rendering of animations to an HP data center at night after its artists go home. And Acxiom Corp. (ACXM ), which analyzes customer interactions for banks, retailers, and credit-card issuers, uses a grid to sift through millions of records per hour.

For those willing to risk trying out a new technology, the rewards can be rich. At most companies, each application typically runs on its own dedicated server. As a result, analysts estimate, organizations use about 15% of their computing power, and they spend far more on servers than needed. By distributing work through grids, companies can save a lot of money. C. Alex Dietz, Acxiom's chief information officer, says he cut his annual data analysis expenses in half by switching from a single $2 million mainframe-style computer to a grid of cheap servers running the Linux operating system. And he can process in four to five days jobs that used to take a month.

Most grids focus on huge number-crunching jobs like Acxiom's, but that's changing. Grids are starting to handle everyday tasks that demand instant answers. Broker Charles Schwab & Co. (SCH ), for instance, has an application that lets investment advisers sit down with clients and size up different strategies, then ask the grid to analyze those strategies and produce scenarios. The answers, which require billions of calculations, come back in seconds.

The dream is to tap into grids as we now plug appliances into outlets. That goal is years off. The technologies are immature, security must be improved, and all the standards aren't yet in place. This will be one of those advances that takes many years to make its full impact felt. But for many corporate computer users, it will be worth the wait.

By Steve Hamm in New York

    Before it's here, it's on the Bloomberg Terminal.