__ An open-source success story. __
When Cobalt Networks founders Vivek Mehra and Mark Orr began building their fledgling server company, they had no idea that open-source software would be a load-bearing part of their strategy. In fact, when their plans began coalescing in late 1996, the term open source hadn't even been coined. But since March '98, when they shipped their first product, the Cobalt team has learned a few things: The efficiencies of using code prefabbed in perhaps the world's smartest lab and - here's the strange part - giving away a good chunk of your own labor are the makings of a real live business.
Today, with headquarters in Silicon Valley and sales offices in the US, Japan, the UK, Germany, and the Netherlands, the company is growing fast. Investors include NTT, Mitsui Comtek, France Telecom, Chase Capital Partners, Tech Fund Capital, Vanguard Venture Partners, and Crystal Internet Venture Fund. While its claim of being the largest supplier of dedicated Net servers to the Mongolian ISP market isn't likely to get Wall Street panting, Cobalt's sales to over 500 US customers between March '98 and March '99 suggest the company is on to something.
Like many start-ups (see "Tour de Source," page 144) in the last two years, Cobalt is on to open source. The elegant, if commercially counterintuitive, model of programming is allowing companies that understand it to reap its benefits and generate profits. Firms such as Cobalt and Cygnus, as well as more venerable companies like IBM, Silicon Graphics, HP, and Compaq, are embracing open source as a means to their individual ends. By focusing their efforts on adding value beyond the basic programming, these open-source-based outfits are daring to explode the old model of hoarding every scrap of your business except what you sell.
Instead, Cobalt and other companies are using strategies of sharing and stockpiling intellectual property in myriad combinations. Some, like Oracle and Informix, are simply jumping on the bandwagon by porting versions of their existing products to run on Linux but keeping their code close to their vests. Others, like IBM and SGI, devote full-time engineers to enhancing Apache and Samba, respectively, and make any resulting new source code freely available. The degree of commitment to the open-source model is a continuum. While companies like Dell merely provide more customer choice by offering Linux preinstalled on new servers or PCs, Netscape has actually taken its own star app and made its source code available for free.
Eventually, creative variations of open-source business practices might include releasing code over time, allowing you to hold on to your precious property long enough to get a running start. Then, when you're poised to ship your next rev and can count on the new version's revenues, you might release the previous source code to the programming community to see what improvements accrue.
The latter model appeals to Cobalt CEO Stephen DeWitt: "This idea has merit; it's how you can keep moving the ball along."
But DeWitt is no ideologue. His company lands somewhere in the broad middle ground when it comes to passions of "the movement." While happily respecting open-source software licensing agreements that require users to share modifications to existing code with the rest of the world, Cobalt still keeps some related, if not integral, developments to itself. "We do feed a lot back to the community, but we're not a charity," points out DeWitt. "We have shareholders." Pragmatism dictates he be community-friendly. Three out of Cobalt's 80 employees are dedicated to developer relations - no small commitment for a company its size. "We're actively courting developers to add value to our systems."
Before joining Cobalt, DeWitt was a vice president and general manager at Cisco. Both Mehra and Orr held senior positions at Apple, Mehra in engineering and Orr in product marketing. The three set out to build a device that took all the tough stuff out of setting up an Internet server. At the time, Net servers were exclusively the domain of corporate IT departments, yet market research suggested that wouldn't be the case for long. According to a Zona Research report published in 1996, Internet servers were forecast to grow into a $2 billion market by the end of '99, while sales of intranet servers were expected to exceed $12 billion. (As of 1998, those markets are estimated at $1.3 and $7.3 billion, respectively.) Cobalt's plan was to take aim at a slice of that bounty: the as-yet-untapped small and medium-size business demographic - typically sans professional systems administrators - which is even now only beginning to realize the Net's value.
What wasn't clear in the early planning stages was how Cobalt was going to achieve its goals. Since the principals had all been through their share of product launches, they took their investigations into the technological underpinnings of their proposed business seriously. Their research into which microprocessor to use and which operating system to build upon took well over two months.
For the motherboard, Cobalt ended up designing its own, based around a RISC chip with ample horsepower for the job; this sub-$40 processor was already being designed into various set-top boxes and routers, guaranteeing economies of scale on the hardware side.
But for software, Linux was Cobalt's tantalizing, if largely untried, solution. It was already a full-featured operating system. It was a stable and well understood piece of technology that had been ported to several different microprocessors and vetted by hundreds of programmers worldwide. Adopting Linux rather than developing its own OS easily saved Cobalt years. In the decade or so it took Microsoft to build Windows NT, or even the eight years it took Linux to evolve, Cobalt's engineers could churn out multiple new products and upgrades. They could concentrate on breaking new ground.
Mehra, Cobalt's VP of products, and Orr, VP of business development, insist that Linux's price - or lack thereof - wasn't its primary appeal. But the numbers speak louder than that: Vendors pay a per-server licensing fee of $850 for Windows NT, while the sticker price for the Cobalt Qube 2's low-end configuration is only $999. The company's highest-priced server, a fully loaded RaQ, is priced at $3,199. Had Cobalt taken the NT route, based on those prices, the Microsoft fee would have sucked anywhere from 27 to 85 percent from the start-up's bottom line.
Another catalyst for choosing Linux, according to Mehra, was the free OS's gathering momentum. Mehra conducted a less-than-systematic survey to measure its popularity. "I went to Computer Literacy," he says, referring to the popular Silicon Valley technical bookstore, "and looked at the number of books about the different operating systems. Linux had the most. It was as scientific as that."
Linux isn't the only open-source software running on Cobalt's hardware. Just as it's the sun around which the vast majority of OSS projects orbit, Linux also is the hub for the Cobalt servers' features. Since initial market research showed that intranet functions were key, the Cobalt team included a discussion-board server as well as print and file servers. As an Internet box, the Cobalt product needed to act like a Web server, a mail server, and a domain name server. For the Web component, Cobalt chose Apache; for the file server, Samba - both are open source. Since these packages, as well as most other open-source efforts, are tried and tested on Linux, porting the software components to Cobalt's hardware was simple.
In addition to being used and improved by programmers worldwide, Apache was adopted, in several senses of the word, by IBM in June 1998. Like Cobalt, Big Blue decided to make Apache's technology integral to its WebSphere product line, but it has also contributed to Apache's further development. Indeed, a good portion of the work required to make Apache run on Windows NT was performed by IBM engineers, whose efforts went straight back into the open-source base. The mainframe giant was under no compulsion to contribute back, given Apache's relaxed license, which permits the company to take the code and run if it wants to. But IBM fed its changes back into the community anyway.
Why? Anyone who's ever worked on a single document with at least one other person can understand the problem. Somebody prints out separate copies of a memo, and each person wanders off to make changes. One of them will get stuck with the odious task of merging the changes together. The longer the two spend working separately or the more people are added to the task, the harder it becomes to synchronize revisions in the end. This problem is very real in the world of software development, and it explains why IBM's approach to Apache is far from altruistic.
Having made the decision to base a part of its business on Apache, a move that allowed IBM to bring a mature piece of technology to market in less than six months, it was a simple matter of enlightened self-interest that IBM followed the open-source model of "giving back." If Big Blue had just absconded with the code, the dreaded synchronization problem would have prevented it from reaping the ongoing improvements made by the rest of the Apache group.
Cobalt understands this dynamic, too, and has shared with the Linux community many of the results of its efforts to port Linux to its hardware. Changes Cobalt made to Samba, a file and print server that to an ordinary networked client PC looks just like a Windows NT server, have also filtered back into its source code. Like SGI, which adopted Samba (and hired one of its core developers) to bridge the gap between its longtime Unix users and its new NT customers, Cobalt has tossed its Samba enhancements back into the pot without getting a penny in return.
Ultimately, Cobalt's use of OSS differs from that of IBM and SGI. Where the two established companies were addressing well-understood needs, Cobalt was making a new market. At the time, it didn't have any direct competitors, and even today, the only way to match the price/performance ratio of a Cobalt server is to build your own and perform all the software integration yourself. This is not for the faint of heart.
Besides, Cobalt's plan wasn't just to take a bunch of freely available software and port it to a custom piece of hardware. Cobalt went on to create its own special - and easy to use - concoction of open-source software, proprietary software, and cheap proprietary hardware.
Cobalt's choice of open-source software for its products has been vindicated many times over. From March '98 to March '99, the upstart introduced seven servers - the Qube, RaQ, CacheQube, CacheRaQ, Qube 2, RaQ2, and NASRAQ, each with different configuration options - and sold them in 55 countries. CEO DeWitt sees the rapid cycle time as another benefit of going with OSS: "We're on our second hardware generation now, introduced ten months after the first Qube and six months after the first RaQ and cache products. We're going to keep up the six-month cycles."
An ordinary mortal nonsysadmin can set up a Cobalt server in 10 to 15 minutes. (Mehra, Orr, and DeWitt know this because they tried it out on all their investors, and only one, who couldn't type, needed help.) With some clever software design under the hood, Cobalt created a Web-based user interface that made setup easy. Hardware configuration consists of plugging the box into an electrical outlet and connecting it to an Ethernet LAN. Everything else happens in a browser.
In addition, though the typical user won't much care, Cobalt ships the full suite of development tools on every box, as well as the open-source code. The user interface is written in the Perl scripting language, which runs only in source form, so though this code is copyrighted and explicitly not open source, Cobalt has no alternative but to ship it as well.
Yet this is not simply a perilous giveaway. By distributing all the individual software parts with its products, Cobalt is inviting value-added resellers to purchase its servers and customize them. Thinking Objects Software, for example, took advantage of the Qube's single, unpopulated PCI slot and stuck an ISDN card in it. The German company's in-house developers ported the card's drivers and some routing software to the Qube and created a user interface that integrated seamlessly with Cobalt's own. A buyer of such a system would never know that the Qube's newfound communications capabilities weren't built-in from the start. The spontaneous appearance of a value-added reseller (Cobalt only learned of Thinking Objects' enhancements post facto) couldn't have happened had the Cobalt Qube been less open. "It was absolutely essential for the Qube to have value added by the channel," says DeWitt. "This was a very, very conscious strategy on our part." Over 70 percent of Cobalt's products are sold through VARs.
But isn't giving away intellectual property to any and all comers corporate suicide? "Building a product that's this easy to use and this optimized for its purpose is not that easy," according to DeWitt. "Our added value is in the middle layers, between the kernel and the hardware platform, and between the OS and the application."
ISPs were among the first to embrace Linux as a viable server operating system, and Cobalt's RaQ offers this market a polished alternative to the painstaking work of constructing servers themselves. As Cobalt's Mehra puts it, "The ISPs tell us, 'It was fun the first few times,' but their focus is on managing customers and growing their companies. They're paying [us] for convenience, ease of use, and integration."
DeWitt adds, "Open source allowed us to compete credibly against established companies, define a new market space, and open that market through new price points."
By mid-February 1999, nearly a year after the launch of the Cobalt Qube, the privately held company won't talk about concrete figures or unit shipments. According to DeWitt, Cobalt sold "multiple thousands, soon to be tens of thousands" of units in its first 11½ months. At a minimum of $1,000 per box, that works out to a conservative $10 million in revenue. The company has grown from 14 to 80 employees and hopes to be profitable before the end of this year.
Cobalt's second-generation hardware, heralded by the arrival of the Qube 2, added more performance and included more extensive server capabilities, yet remained at the $999 introductory price of the original Qube. "We intend to add more functionality at a constant price point," says DeWitt. "That's the way this category is going to go for some time."
Over the next 12 months, Cobalt aims to maintain its pace of new product development - in most cases multiple products developed in parallel. It's also eyeing an IPO and planning further international expansion. DeWitt's looking closely at Russia, China, and Japan. "We have the opportunity," he says, "to become a standard in markets like Japan due to early investment there."
Cobalt's founding trio is careful to point out that open source won't work for everyone, however. In other words, don't try this at home. "The model has demonstrated that it can evolve more quickly and with higher quality than the proprietary model in the Internet space, but not necessarily in other application spaces," says Mehra. DeWitt points out that "the road is already littered with open-source failures." Corel, which built a hardware division dedicated to producing custom Linux machines, sold it off earlier this year.
And DeWitt has all the perspective of a CEO staring at his bottom line. Despite the obvious successes of Cobalt's first year out of the gate, he tempers the euphoria that's spreading among open-source entrepreneurs thinking of taking their creations to market.
"Open source is a great foundation, but you still have to build a solution that grows and defines a market," he says. After a year and a half, he's only begun. "It's not hard to build a better mousetrap, but it is hard to build a business."
