Sep 28, 2003 (08:09 PM EDT)
The Guts To Say 'Go'
Read the Original Article at InformationWeek
Daniel Kaberon had a reputation. If anyone started getting too excited about grid computing, Kaberon could be counted on to throw a big bucket of cold water on it.
"I was a loud voice saying it makes sense if you're looking for extraterrestrials or figuring out the number of vowels in the New York phone book," says Kaberon, director of computer resource management at Hewitt Associates LLC, a human-resources consulting and outsourcing company that manages benefits for one in 20 Americans. "I didn't see how it worked in a business process."
Unproven technology is decidedly out of style at many companies. With IT budgets so tight, many executives play it safe with the tried and true and stay away from the bleeding edge. Money and time to experiment with new technologies have to be hidden in budgets, if they can be found at all.
Yet some companies--and Hewitt is one of them--still believe that being early adopters of technology gives them a competitive business edge. In fact, Hewitt thinks that's the only way to get competitive advantage out of business technology: get in early in the adoption cycle, or stay away until all the bugs are ironed out. "We want to be at the very beginning or the very end," says CIO Perry Cliburn.
The grid-computing project shows that philosophy in action.
Hewitt manages HR benefits for 16 million people, processing more than 69 million online transactions per week for companies such as Sony Electronics Inc. and Johnson & Johnson. Hewitt had a problem, though, that centered on one calculation: Figuring an individual's pension benefits based on his or her company's specific plan. Hewitt's traditional approach was to use a mainframe to do that calculation, but the Internet was making that too expensive. Pre-Internet, employees might have gotten a printed individual report when they retired and maybe an update when they were 50 or 55. Now HR managers expect instant access to that number online, as often as they want.
Most pension calculations are easy: Harry worked 17 years for XYZ company under the same pension rules. You need a calculator to do it, not an IBM Parallel-Sysplex mainframe. But every once in a while there's a humdinger: Harry started with XYZ, which was bought by ABC, which merged with CYA, which changed its pension rules five years after the deal, after which Harry took a two-year leave of absence, then came back under the management plan, not the union plan. Calculating the 1% of cases like Harry's consumed 25% of the mainframe-processing power Hewitt used for pension calculations. Having expensive mainframe capacity waiting around for the Harrylike 1% to come through the system wasn't practical. Cliburn told his team to find a better way.
Finding a better way is an obsession with Hewitt's IT group under Cliburn and chief technology officer Sanjiv Anand, who has championed the company's tech strategy for more than a decade. Two-thirds of Hewitt's revenue comes from HR outsourcing services, which handle everything from managing health-insurance benefits to cutting payroll checks for companies that can have tens of thousands of employees. Client companies consider information about employees their data, so if they want access to it, Hewitt needs to figure out ways to deliver it while still making money.
Hewitt considers the systems that run its outsourcing business one of its main competitive advantages. When prospective clients come to its corporate campus in the Chicago suburb of Lincolnshire, they're treated to the "Heartbeat" video--sort of an XML meets MTV effort that showcases Hewitt's early adoption of tools, such as Web services and Parallel-Sysplex mainframe clusters, as key parts of its marketing message.
Yet Hewitt doesn't hold a single patent related to that technology. There's nothing stopping rivals from copying what Hewitt has built. Instead, Anand's philosophy depends on identifying and executing on the next technology-driven innovation that will let the company maintain market leadership.
Speed and innovation don't always lead to glory, of course. Take Hewitt's Sageo project. In retrospect, Sageo was so January 2000: launch a standalone, Internet-based spin-off offering standardized health and welfare benefits for companies with less-complex needs. Sageo brought in $25 million in revenue over its two-plus years but had a $91 million cumulative operating loss. Beginning in September 2001, the Sageo Web site was integrated into Hewitt's larger system, its software written off as a $26 million onetime charge, and its clients integrated into Hewitt operations.
But Hewitt has far more often succeeded as an aggressive early adopter of technology--like component architecture before it was cool, and before there was Java or the Web. In the early 1990s, the company began planning a technology platform to offer integrated delivery of all of a client company's employee benefits from one data source and code base. The IT architects of that system, which was unveiled in 1995 and called Total Benefits Administration, or TBA, knew they would have to efficiently reuse that platform for every new customer. So they wrote the system in Cobol on a mainframe in a way that's similar to what people do today for Web services. "The only object-oriented language we had was Cobol," says Tim Hilgenberg, chief technology strategist for applications. "It looks like a container system like J2EE."
Hewitt has since added access to that system via telephone voice response, the Web, internal customer-service reps, and, most recently, Web services, without changing the underlying TBA system. Now the TBA system exchanges data using XML with employers' PeopleSoft Inc. systems or insurance companies' transaction systems. "Every time we've been asked to add a new channel, we've been able to do that," Hilgenberg says.
That's where grid computing comes back into the picture.
Hewitt's IT staff got serious about a grid option around January of this year, which set up a race against time. In the HR world, the calendar turns around open enrollment, that one and only time period in the fall when everyone can change their insurance plans. If this were retail, open enrollment would be Christmas, Halloween, and back-to-school sales rolled into one.
Hewitt staff teamed with people from IBM's Design Center in Poughkeepsie, N.Y., to create the grid. The technical challenge was moving the code that ran the calculation, written in SmallTalk, from the mainframe to a Linux operating system. By May, they had a version ready for production. "It was a fast-moving train," says Paul Wanish, the IBM grid and systems architect on the project.
Grid computing is usually thought of as using idle computing resources, distributing a task among PCs with unused processing power. Grid got famous for the Search for Extraterrestrial Intelligence project, in which individuals donate idle processing time on their PCs to analyze radio-telescope data (see sidebar, "Grid Computing: The Scoop On How It Works").
Two big questions loomed over the tests: Would the grid results be accurate, and would they be as fast? Hilgenberg, Hewitt's applications strategist, was sure they wouldn't be accurate. The tests showed it was accurate, 0.2 seconds slower on the easy calculations, and 10 to 15 seconds faster on the tough ones. "I went into it saying, 'This isn't going to fly,'" Hilgenberg says. "I came back, and colleagues were saying, 'What happened to Tim out in Poughkeepsie?'"
Wanish of IBM thinks other people will be going through their own grid-skeptic conversions soon, the way Kaberon and Hilgenberg did. He thinks IT architects will increasingly think of business-computing work in parts--dividing them among the functions that need a pricey mainframe and those that can be carved off and done in less-expensive ways. And just as Hewitt had never before used Linux, he predicts more companies will open up to that possibility.
Tony Bishop, the chief business architect of DataSynapse, says Hewitt is a visionary in using grid. Most commercial companies using grid--and there aren't that many to start with--aren't using it in a dynamic flow of a business process as Hewitt is. And since this uses a standardized means of Java middleware and Simple Object Access Protocol messaging to route part of the mainframe's work to a cheaper alternative, it opens up the possibility of doing more on the grid. "They now have a way to migrate the right type of request to the right type of computing source," Bishop says. "Since they've built it through a Web service, open program, they've allowed this mainframe to talk to the distributed world." He predicts commercial grid computing will evolve in this direction over the next two years.
Hewitt's grid function is working, serving select clients who are aware of the new technology being used, and the IT staff is slowly increasing the load on it. It has worked without surprises so far.
Hewitt execs know the greatest payoff won't come from cutting the cost of one pension calculation. This is not only the first grid effort the company undertook but also the first Linux project. None of the Hewitt staff sees grid computing replacing the mainframe environment for massive transactions. It might have uses in some of the calculation-intensive work done by Hewitt actuaries. But like any true business-technology innovator, Hewitt's staff knows that the best idea is the one they can't even imagine yet. Says Hilgenberg: "If we build a competency around grid, we will start to see opportunities we don't now."