A challenge for bug exterminators
Five-year redesign of Microsoft Windows and Office comes down to a final scramble to get the programs to users.
The principle behind the curve--that 80 percent of the consequences come from 20 percent of the causes--is rooted in a 19th-century observation about the distribution of wealth. But it also illustrates the challenge for the builders of the next generation of Windows and Office, the world's largest-selling software packages.
As they scramble to get the programs to users by the end of the year, the equation is a simple one: making software reliable for most personal computer users is relatively easy; it is another matter, in a PC universe with tens of thousands of peripherals and software applications, to defeat the remaining bugs that cause significant problems for some users.
The effort to overhaul the Windows operating system, originally code-named Longhorn and since renamed Vista, was meant to offer a transformation to a new software foundation. But several ambitious initiatives failed to materialize in time, and the project started over from scratch three years ago. The result is more an evolutionary shift, focusing on visual modernization and ease of use.
Still, the company is within a month of completing work on new versions of both Windows and Office, having apparently overcome technical hurdles that as recently as August seemed to signal a quagmire.
"It looked bleak; it was a slog, but in the end this was a technical problem, and there was a turning point," said Bharat Shyam, 37, a computer scientist who is director of Windows program management. "We've confounded the analysts and the press."
As October arrived, a vote of confidence came from Wall Street when a Goldman Sachs analyst, Richard G. Sherlund, wrote that he expected the product to be introduced on time. "The Vista development organization has made rapid progress delivering improvements to Vista's performance, reliability, and compatibility," he said.
(On Friday, the company released what it said would be the final test version of Vista, named Release Candidate 2. If the response from testers is positive, the software will go into production by the end of the month.)
The debugging process has been urgent, with Microsoft scheduled to introduce Windows Vista and Office 2007 to corporate customers by the end of the year, and to home users early next year.
This coordinated introduction is a multibillion-dollar proposition for Microsoft, which has Windows running on some 845 million computers worldwide and Office on more than 450 million, according to the market research firm Gartner.
Indeed, it was the vast scale of the Windows testing program that saved the software development projects. Over the summer, the company began an extraordinary bug-tracking effort, abetted by volunteers and corporate partners who ran free copies of both Windows and Office designed to send data detailing each crash back to Microsoft computers.
The Office package, for example, has been tested by more than 3.5 million users; last month alone, more than 700,000 PCs were running the software, generating more than 46 million separate work sessions. At Microsoft, 53,000 employee computers are running test versions.
Vista has also been tested extensively. More than half a million computer users have installed Vista test software, and 450,000 of the systems have sent crash data back to Microsoft.
Such data supplements the company's own testing in a center for Office referred to as the Big Button Room, for the array of switches, lights and other apparatus that fill the space. (A similar Vista room has a less interesting name--Windows Test Technologies.)
This is where special software automatically exercises programs rapidly while looking for errors.
The testing effort for Windows Vista has been led by Mario Garzia, Microsoft's director of Windows reliability. A former Bell Labs software engineer, Garzia says the complexity of the Vista and Office effort dwarfs anything he undertook for the nation's telephone network.
"Everything is easy if you do it for a limited number of things," he said. "When I was at Bell Labs, the problems were complex, but nothing compared to this."
The test data from the second beta release of Vista alone generated 5.5 petabytes of information--the equivalent of the storage capacity of 690,000 home PCs.The resulting complexity can be seen in the dance that has gone on in recent months between Microsoft's designers and its partners, who have been tailoring software and hardware to work with Vista.
On Sept. 1, for example, Microsoft released a version of Vista called Release Candidate 1 to a large group of outside testers, hoping to take advantage of their free time over the Labor Day weekend.
Immediately, Garzia recalled, a wave of crash data fed back to Microsoft disclosed a newly introduced bug that had been created by incompatibility with a software module (referred to as a device driver) written by a partner company.
That company was alerted to the problem, and a remedy was transmitted directly to the testers' computers over the Internet within four days--a vast improvement in the gap between detection and repair, he said.
Despite the impending commercial arrival of the two software projects--which between them have involved the labors of more than 5,000 programmers and testers here--there is still uncertainty in the industry about how long it will take for Vista in particular to gain acceptance.
"We've been impressed with the progress, and they deserve a lot of credit," said David Smith, a Gartner vice president, but that does not mean that Windows Vista will soon be in standard workplace use. Its deployment on a significant scale will not begin at most companies until 2008, Smith said.
Microsoft executives contend that such calculations are overly conservative, and they have been making the case that the use of Vista could pay for itself in saved labor and related costs in less than a year.
A more fundamental question for the industry is whether Vista will represent a new era for computing or be the last great push of the current epoch.
While Microsoft's co-founder and chairman, Bill Gates, was able to turn his company abruptly in the mid-1990s to respond to the challenge posed by Netscape, Microsoft has proved less effective in blunting a similar challenge to its dominance from Google.
Moreover, the rise of Google and other companies moving toward Internet-based software development raises doubts about the value of giant efforts like Windows and Office, which can take more than five years.
Eric Schmidt, chief executive of Google, has said he believes that the rise of advertising-supported Web services will increasingly undercut Microsoft's software development model--using a proprietary software development system and selling shrink-wrapped applications.
In an internal company memo titled "Don't Bet Against the Internet," he wrote recently, "Almost no pure PC software companies are left (all is on the Internet), most proprietary standards (I'm thinking of Exchange e-mail and file systems protocols from Microsoft) are under attack from open protocols gaining share rapidly on the Internet."
The larger struggle has had little influence on Ben Canning, who began his career at Microsoft testing software nine years ago after getting a graduate degree in philosophy from Reed College.
Rather, his days are consumed with working his way down that whiteboard curve.
Canning acknowledges that his degree prepared him for little beyond teaching philosophy--with the possible exception of finding and killing bugs in software, because philosophers are trained to analyze and solve particularly hard logical problems. For the last few months, his mind has been focused on the hard problems at the end of the curve.
"If you look at the mean time to crash for most Office customers, it's very high," he said. "There is a small minority that crash all the time, and they hate us, and we want to help."