YOU HEAR A LOT about garage startups in high-tech, says Edward Scott Jr., gray chest hairs

poking out through his open-necked Hawaiian shirt.

Were a condo startup.

Its true. Scott, 59, helped start Mountain View, Calif.-based BEA Systems in partnership with fellow Sun Microsystems alumni Alfred Chuang, 36, and William Coleman III, 50, in 1995. They held early meetings in a condominium.

At an average age of 47, they were over the hill by Silicon Valley standards, and there was nothing revolutionary about their software.

Middleware, as it is called, is a software-based traffic cop of sorts that makes sure messages between computers reach their destinations in an orderly fashion. If invisible, it is, however, ubiquitous. When installed in all the computers in a network, it allows different kinds of machines and programs to talk to each other.

Middleware probably wont appear in a Sandra Bullock movie anytime soon. But the lack of sizzle didnt stop investment firm E.M. Warburg, Pincus & Co. from handing the three men $50 million for half of their company in the summer of 1995, when it was still not much more than a 325-page business plan.

Two-and-a-half years later, following an initial public offering in April and a secondary offering in July that raised $150 million, Warburg, Pincus investment is worth about $500 million.

Yes, old fogies can deliver growth. BEA (after the founders: Bill, Ed and Alfred) had sales of $131 million in the 12 months to October 1997, up from $41 million the year before.

The idea that became BEA germinated in 1993. After seven years at Sun Microsystems, Coleman took a six-month leave of absence, most of which he spent at his condominium in Aspen. Between ski runs, Coleman ruminated on what a startup might do to capitalize on a trend sweeping through the worlds largest corporations: the shift of data processing from a few large mainframe computers to networks of small, cheap, server machines like those sold by Sun.

Sun itself was then in the throes of shifting its most critical, time-sensitive data processing from old, dependable IBM mainframes to its own servers. It was in a dilemma: Server-based systems are far cheaper and more versatile than mainframes — but are more likely to break down. So data addicts like banks and airlines are making the switch gradually, to shake the bugs out. That creates another problem:
integrating old mainframe-based systems with those running on new servers.

Upon his return to Sun, Coleman cleverly signed on to head Sun Integration, the companys mainframe-to-server migration project.

During a dinner party at his house in 1994, Coleman approached Scott and asked him to step into his home office for a chat. The two had been friends since 1984, when they founded Dest Systems in Milpitas, Calif.

Just a year later the two watched Dest disintegrate when the venture capital behind it dried up. After that they worked together at Sun Microsystems before Scott left to head sales and marketing at Pyramid Technology Corp.

Coleman had already told his bosses that he planned to leave Sun at the end of year to start his own company. Now he told Scott what he had in mind: Come up with a software package that would let mainframes coexist with client-server systems. Theyd have to move fast to beat server vendors like IBM, Digital Equipment Corp. and Hewlett-Packard to the punch, which meant that they would have to buy technology rather than build it.

That was where Scotts dealmaking skills would come in. Known for snatching a huge contract to supply servers to the Internal Revenue Service away from IBM while he was at Pyramid (FORBES, Aug. 19, 1991), the ruddy, back-slapping Scott would engineer the purchase of promising small companies that were working on the problem and strike licensing deals with bigger companies. Scott, who was negotiating the sale of Pyramid to electronics giant Siemens Nixdorf at the time, signed on.

Now all they needed was a smart engineer. Two weeks later, Hong Kong-
born Alfred Chuang, Colemans chief scientist on the Sun Integration project, walked into his office and announced that he intended to leave Sun to take a job as the head designer of information systems at Charles Schwab & Co.

Although Sun had forbidden him to recruit from the companys ranks for his startup, Coleman figured Chuang was leaving Sun one way or the other. Stay here until the end of the year, Coleman told him, then join my startup.

Colemans concept was music to Chuangs ears. His dream: to come up with a system that would fuse every computer attached to a given network into a cohesive entity — a system where the network, as the Sun slogan goes, is the computer. Rather than depending on a central mainframe or server for accessing data (see box), software on the system would seek out the desired information wherever it sat on the network, process it and deliver it instantaneously. This architecture would allow companies to migrate to smaller servers without having to throw out their mainframes.

I might have been able to build [an answer] from the ground up, Chuang says. But it would have taken three years. We didnt have that kind of time.

Another problem was that no large corporation was likely to risk its business on unproven technology. I cant invent maturity, Chuang told Coleman. They had to build their system around something that was already in use. They settled on a program known as Tuxedo, developed in 1983 by programmers at Bell Labs to enable large numbers of users to simultaneously access and manipulate a database on a mainframe. Much like the Unix operating system that runs most network servers, Tuxedo had dribbled out of Bell Labs into the commercial world. There were several variations of it being sold by a handful of small companies, plus Digital and Novell.

Armed with the $50 million war chest from Warburg, Pincus as well as some bank credit, Scott went to work, buying up the smaller Tuxedo-based companies and trying to pry the source code away from the bigger players. It took Scott six months to convince Novell that it was better off selling Tuxedo to BEA than trying to integrate the complex software into its own product line. The price: $90 million. BEA also bought talent. The guy who wrote the first line of Tuxedo code works for me, crows Chuang.

With his hands on the source code — the master copy, so to speak, of a computer program — Chuang set about weaving the various versions of Tuxedo into a single program. He also added modern enhancements, including a module written in Suns Java language that lets users manipulate mainframe-based databases over the Internet.

Happy to end — gradually — their dependence on mainframes, corporations like Fidelity Investments, Federal Express, AT&T and Northwest Airlines have installed BEAs software. But BEAs hardest work in selling is still ahead of it. Microsoft, IBM and Oracle are all working on competing systems. Coleman says confidently, By the time Microsoft is ready with theirs, well have a defendable leadership position.