Sun's Computing 'Utility' Is More Like Equipment RentalSun's Computing 'Utility' Is More Like Equipment Rental

Most research-intensive industries, such as financial services and pharmaceuticals, want to mitigate risk. They can do so if they move to variable rather than fixed costs: As their businesses shrink or grow, so will operating costs. For that reason, the buzz around "utility" or "on-demand" computing has grown louder. Sun's recent launch of its utility computing service would seem to be the long-awaited manifestation of this talk, but it's not.

"True utility computing would allow us to dynamically accommodate computing needs for multiple departments across multiple lines of business, without additional hardware purchases or maintenance costs," says Whit Stockwell, CTO of Genalytics, which uses genetic algorithms to expedite modeling for marketing and risk management. While Sun's "utility" launch piqued his interest — Genalytics develops in Java — Stockwell didn't buy in because, he says, utility computing has to mean more than just adding capacity for peak usage times.

Sometimes claims of "utility computing," such as Sun's, turn out to be nothing more than cluster computing. True utility computing requires the ability to respond quickly to need. That entails having grid software that's sufficiently intelligent to match customer service-level agreements with spikes in computing needs, so computational power is sought out automatically and without advance notice.

"Utility computing isn't a new technology, but a business model, for which grid software is the foundation," explains Bob Boettcher, VP of financial services for Platform Computing, a leading grid software provider. Grid computing is now only at a point of "scheduled on-demand"; companies need to schedule ahead of time the purchase of extra CPUs for anticipated spikes, such as at the end of billing or reporting cycles.

"Renting a grid from Sun with a couple hours notice is just not possible now," says Nick Gall, SVP, MetaGroup. "I believe the days of true utility computing, where companies run their traditional data center applications on a grid, are a long way off."

Indeed, many view grid as an evolutionary step toward utility computing. Most companies will need to wet their feet by putting one application on an internally hosted grid to expedite computation times. That test would be limited to a single department. After the company is comfortable, it could begin moving multiple applications and lines of business onto the grid. "When multiple departments are sharing resources across multiple lines of business, then the IT department begins to operate in a true utility fashion," says Boettcher. "Utility would enable all lines of business to use the same computers, with IT charging on a variable basis." The next evolutionary step could then be offloading to a host.

It all depends on whether the economics of utility computing will ultimately make sense. Sun will bolster its data centers and capabilities to support utility computing. However, Sun's strategy of provisioning a dedicated grid of clusters to each customer might not pay off. It would be better to have a flexible infrastructure that can share resources to accommodate customers concurrently. "In other words," Boettcher says, they might need to "sell the same CPU multiple times to different business units across different customers" for economic viability.