October 30, 2003 3:30 PM PST

Sun expands utility pricing technique

Sun Microsystems is expanding a program to sell computing horsepower in a way that's similar to how power companies sell electricity.

The Santa Clara, Calif.-based server seller has broken computing power down into a basic chunk: the Sun power unit. It's the amount of work a single Sun UltraSparc III processor can perform in a second, and customers can pay for computing work based on how many Sun power units they consume, said Bill Mooz, Sun's senior director of utility computing.


Get Up to Speed on...
Utility computing
Get the latest headlines and
company-specific news in our
expanded GUTS section.


With "every billing cycle, we will produce a report that will show the amount of Sun power units consumed," Mooz said. "Based on that, there will be an itemized bill." Sun has a separate utility pricing measurement for storage capacity a customer uses, he added.

Sun's utility pricing plan, which began in April, often involves partners that house Sun equipment customers can pay to use. The company's first partner, Affiliated Computer Services (ACS), signed on in September, with SchlumbergerSema joining earlier in October. This week, Sun added another partner, CGI Group, which will offer utility pricing specifically for biotechnology research companies, including Genome Quebec.

Utility computing is catching on, but in some ways, it's recycling an old philosophy, "time sharing," which companies used widely in the 1960s, when computers were scarce and several companies would share one.

"This is one of the classic examples of 'everything old is new again,'" Illuminata analyst Gordon Haff said. "The reason it used to be very popular was that even a billion-dollar company wouldn't own their own mainframe. They were expensive and required expensive little gnomes to keep them running. Then computers got cheap, and everybody bought their own, and it stopped being popular."

Utility pricing, while still rare, is experiencing a renaissance, Haff said. One reason is that it's harder these days to know how much computing power you'll need.

"In general, there's less predictability about business--so if you can shove off some of the risk and uncertainty by paying for what you use, even if it's at a bit higher rate, you're still often better off," he said.

Pooling resources
The idea behind utility computing--which sports labels such as "on-demand computing" at IBM, and "adaptive enterprise" at Hewlett-Packard--is that computer users can save money by pooling together computing resources previously separate. A key issue in uniting these "silos" of computing gear is figuring out which jobs are consuming the computing power and which jobs take priority.

For example, a company could choose to merge computing systems for customer support, accounting and online sales. Customer support could have steady computing demand, while computer usage for accounting could spike at the end of every quarter, with online sales activity surging during the holiday shopping season. Ideally, with a utility computing infrastructure, these differing jobs can be shifted among various computing systems to make sure that equipment that's been paid for isn't idle, high-priority needs are met and that the appropriate departments are billed for what they use.

The future of computing will separate software from hardware so that programs can be quickly moved to appropriate equipment and processing power can be devoted to the most important jobs, Jonathan Schwartz, head of Sun's software group, said in an earlier interview. "You want to decouple the execution of the application from the box," he said. "You will divide hardware from software."

A centralized utility computing site such as Affiliated Computer Services can share resources among several clients. It's analogous to a phone system, in which overall demand stays roughly level, even though one family may make a lot of phone calls during one part of the day and not use the phone at all during other times.

Utility pricing can let a company deal with occasional surges in computing needs without having to pay for all the hardware and software that would rest idle during other times, Sun's Mooz said. But there are cases when utility pricing doesn't make sense.

"The place I don't think a customer will do very well with utility-based pricing is where they have high, predictable and stable levels of utilization. They may be better off buying or leasing the product," he said.

Competitors active
Sun's top two rivals have similar plans under way.

HP's lab researchers are working on a technology called the "computon," said Nick van der Zweep, director of utility computing at HP. The computon is an amalgam of processing power, network bandwidth and storage capacity.

Ultimately, HP hopes that companies will sell spare computons on the Internet, van der Zweep said, not unlike a currency.

He recognizes that there are difficulties packaging different types of computing capacity as one computon, though. "You may want to just store something. But if you want storage, you don't want to be given compute power," he said.

HP has various utilitylike technologies that have shipped in about 10,000 products sold so far, van der Zweep said. For example, a customer can pay for a server based on how much computing capacity is used monthly. HP's terms specify that as soon as a customer has paid more in utility pricing than the total cost of a 36-month lease, the customer no longer has to pay HP anything. That's to reassure customers who fear that unpredictable usage may mean that the arrangement proves more expensive than leasing, the company said.

IBM also has some utility pricing plans. For example, with its so-called on-off capacity on-demand technology, server customers can pay to fire up extra computing capacity temporarily.

IBM Global Services sells access to centrally located servers with a utility pricing model. In one service, for using Linux on a mainframe, Big Blue charges $300 per month for each "service unit," a measurement of computing capacity equivalent to about a third of the horsepower of a single-processor Intel server.

In addition, Big Blue in September expanded the program from Linux on its zSeries mainframes to include the AIX version of Unix on its pSeries servers, the OS/400 operating system on iSeries servers, and Windows on xSeries, IBM spokesman James Sciales said.

CNET News.com's Michael Kanellos contributed to this report.

 

Join the conversation

Add your comment

The posting of advertisements, profanity, or personal attacks is prohibited. Click here to review our Terms of Use.

What's Hot

Discussions

Shared

RSS Feeds

Add headlines from CNET News to your homepage or feedreader.