If Sun's acquisition, in 1996, of Cray's Unix server business for under $100m was one of the best acquisitions in the history of the computer industry, then its purchase of Cobalt Networks four years later for over $2bn (£1.12bn) was surely one of the worst.
Most of us, I'm sure, would not find it easy to blow a couple of billion dollars. Half a dozen houses dotted around the world, a private yacht and even a Gulfstream jet would put only a small dent in your bank balance. Going back a couple of years, you could of course buy a dot-com, but you'd probably want to pick something with no tangible source of income, no real product to speak of, and founders who know more about marketing than they do about technology.
You would probably have avoided something Cobalt Networks, the company that kicked off the appliance server revolution, like the plague. There would be no way that a company whose products were bought by the bucket-load and enjoyed a loyalty rarely seen in the world of IT outside of Apple would ever be a bad financial bet.
In the late 90s, ISPs around the world were filling their racks with Cobalt's pizza-box-shaped appliances. These servers were ideal for ISPs hosting a growing number of Web sites. In a world that was still being sold monolithic servers with colossal service contracts from the likes of Sun, Cobalt Raq servers were relatively cheap to buy, and supremely easy to set up. Once a user account had been created, the user could just be left to get on with it: typically they would have everything they needed to run email accounts and create a database-driven site using PHP, MySQL and, latterly, even Microsoft's Active Server Pages, thanks to the Chilisoft technology that Cobalt had bought. They even had all the tools at hand to monitor the health of their system, set disk space allocation for different accounts and monitor how full that allocation was getting.
It's little wonder that Sun Microsystems, which was becoming increasingly worried by these new server appliances, started making approaches. Ed Zander, who was then Sun's chairman, was particularly interested in the appliance idea and, when I met him in the summer of 2000, was acutely aware of the threat that Cobalt posed to Sun's business. At the time, remember, Sun was positioning itself as "the dot in .com", largely on the back of the high-end servers that it acquired with the Cray purchase. In fact, while Sun probably had the larger sites tied up, Cobalt almost certainly had a lot more to do with the boom in small Web sites.
Indeed, Cobalt enjoyed an enthusiastic following among ISPs, resellers and even end users (often small companies and start-ups). The partner events were well attended. User groups sprang up: the Japan-based Cobalt Users Group even promotes on its Web site Cobalt-branded beach sandals, which leave the Cobalt logo imprinted in the sand.
So when, in the autumn of 2000, Sun announced that it would acquire Cobalt for $2.2bn, the only thing that came as any surprise was the price. By any measure, Stephen DeWitt, Cobalt's chief executive at the time, drove a good deal for the Cobalt shareholders.
If nothing else, the price indicated the faith that some at Sun had in the potential of server appliances. But did Scott McNealy share this faith? I suspect not, given that the Cobalt acquisition was succeeded by Zander's departure which, in turn, has now of course been followed by the closure of the Cobalt business unit.
So what went wrong?
Well, for a start, Cobalt appliances were far from faultless, both in terms of hardware and of software. If you want to secure areas of your Website using .htaccess, then you had to manually edit Apache's misconfigured (on the Raq) httpd.conf file -- a job that is not for the faint of heart, and certainly not for the Cobalt target market. The fans were notoriously unreliable and for years almost impossible to replace. A Cobalt Raq that I owned died completely one day. Re-installing the operating system, which suddenly had a major problem with the two RAID disks, entailed going out to buy a new network card because the Raq's build of Linux did not support the one I had. But the .htaccess problem could be edited away, fans are now available (on eBay), and network cards cost a pittance.
It was not the hardware or the software that doomed Cobalt.
First, there was a clash of cultures: in any case where a huge, established company buys a small, dynamic outfit who are out there doing their own thing and making a success of it, there are going to be problems. Many disgruntled Cobalt engineers left Sun in the months following the acquisition, and more were later made redundant.
Then there was the collapse of the dot-com boom, which began even as the ink was drying on the contract. As the customers of ISPs, and then ISPs themselves, went out of business, demand for Cobalt servers stalled. And nobody from Sun ever really appeared to believe themselves when they said that the Cobalt brand could coexist within Sun, or that a big Unix vendor could support Linux-based server appliances. As the Cobalt logo slowly shrank on the outside of the boxes, inside Sun played around with a couple of Linux distributions. Even Sun's Cray acquisition meant it was buying a business that licensed Sun SPARC processors and the Solaris operating system. With Cobalt, Sun suddenly found itself supporting Linux on Intel (and, with the Qube, MIPs) processors.
And, finally, server appliances have still not fulfilled the dream of the Cobalt founders: that of revolutionising the way software is sold. They have certainly made their mark on the edge of the network, where, according to analyst firm IDC, companies are increasingly turning to the appliance model to buy security software. Storage appliances are also taking off big time. But mainstream application appliances? The closest the industry ever really came was database appliances such as Oracle's Raw Iron and, more recently, devices such as the MySQL appliance sold by Pogo Linux. Certainly Sun, which has plenty of software in its stable, never chose to sell its applications pre-loaded on a Cobalt appliance. After the departure of Zander, it seems, the company simply lacked the commitment to follow through with the appliance idea.
All these forces combined to wipe out Sun's $2bn-plus investment. The announcement of the release of the Cobalt custom user interface and back-end code under open-source licence is proof, if any were needed after Sun put its one remaining Cobalt server out to pasture last month, that that the company has no further use for the Cobalt brand or technology.
Others however, might have a use for the technology if not the brand. The release of the software under open-source licences is great news for Cobalt fans, who will be waiting for someone to combine it with cheap hardware (there is plenty around) and create a successor to the Raq.
But for Sun, it means there is nothing left of Cobalt except perhaps for the Chilisoft software. That aside, Zander and NcNealy might as well have taken that $2bn and bought footprints in the sand.