Can There Be Only Two Linuxes?

Published: April 18, 2006

by Timothy Prickett Morgan

A week before the spring LinuxWorld tradeshow got underway in Boston in March, Novell's chairman and CEO, Jack Messman, was quoted in the trade press as saying that eventually, when all the dust clears, there would only be two commercial suppliers of the Linux operating system: Red Hat and Novell.

It is tempting to think that markets simplify so quickly and easily, particularly when you are one of the companies that is prognosticating who the survivors will be in any market and you want to assure people that you will be one of them. But Linux, and indeed any open source software project that is backed, fronted, or both by a commercial entity, is not playing by the rules that have prevailed for the first four decades of the information technology business. And that means that such consolidation as Messman is talking about may not be as inevitable as many think and some desire.

Having said that, the history of business does seem to suggest that markets start out with many innovators and then condense into a few dominant players. There are usually fewer than a half-dozen dominant players in any market, whether you are talking about making cars, movies, televisions, appliances, food, or what have you. In the old days before massive globalization, you would see such consolidation on a national scale, but now consolidation is global. This was enabled in part because of cheap oil to move goods, cheap telecommunications and computing to coordinate global efforts, and the move by governments to lower trade barriers and prop up international trade as a means to boost their own economies. We can debate how successful this has been, but the fact remains that consolidation has continued apace in any market you can think of, and now, we are at a point where a few global players dominate most industries.

Let's look at an interesting example of such consolidation. Back in the dawning of the mainframe age in the 1960s, by IBM's own assessments of the mainframe market, only itself, Sperry Rand, Scientific Data Systems, and Control Data were profitable companies engaged in making and selling mainframes, while IBM figured that Burroughs, General Electric, Honeywell, NCR, RCA, and Singer were all losing money. It wasn't long before RCA sold to Sperry Rand, SDS sold to Xerox, GE sold to Honeywell, which then bought out Xerox a bit later, and Singer sold to the British computer maker ICL. Then, Amdahl jumped into the market with clone IBM boxes, and got backing from Fujitsu to try to take on Big Blue. Hitachi and NEC had created, with Fujitsu, an indigenous mainframe market in Japan, Siemens ruled Germany, and Bull ruled France; all of these companies had to share with the local IBM affiliates--if having the second position in your home market can be called sharing.

But in the late 1980s, the price/performance curve pushed Sperry and Burroughs together to make Unisys, which just last year gave up all server design except for its mainframes to NEC; ICL, Amdahl, and then Siemens allied themselves with Fujitsu over the 1980s and 1990s, while Bull struggles along alone, supporting its mainframe environments on Itanium gear, just like Unisys does with its MCP environment. Still, if you are talking about supporting COBOL applications, what you really have these days in terms of mainframes is a choice of two boxes: IBM's System z and Unisys' Clearpath. And yes, you can run COBOL on Unix, Windows, and Linux boxes, using software from Micro Focus, Acucorp, or Fujitsu, but most mainframe shops have more than COBOL that makes the mainframe appealing. They have expertise in making these machines work well, which is why they stick with them even though they are terribly expensive from a cash outlay point of view.

In the 1970s, the proprietary minicomputer market similarly had a diverse collection of players that has basically consolidated down to two players: IBM and Hewlett-Packard. And the more diverse Unix market that exploded onto the scene in the 1980s, which gave birth to Sun Microsystems, Silicon Graphics, and a host of other new players in the server space as well as giving incumbents like IBM, Digital Equipment, HP, and others a new line of business to sell, has basically consolidated down to three players--IBM, Sun, and HP--locked in a three-way battle with nearly equal pieces of the Unix market for the past five years. (IBM is growing fast, HP is starting to grow again, and Sun is shrinking but stabilizing.)

So will this crunch happen to Linux and other open source software as well? Is it a desirable result if it is not inevitable, or is consolidation something that the industry should try to avoid?

The major IT suppliers clearly do not want to support more than two Linuxes, because this simplifies their sales and certification efforts. One open source Linux distro, middleware stack, or database cannot cover all needs, but given two vendors of reasonable breadth and depth--with lots of vendors certifying the software on their hardware--you can get decent coverage on the hardware below and the software above. It is easy to understand why IBM, HP, Dell, Sun, and Fujitsu-Siemens, the five dominant server makers, want Red Hat and Novell to be the only Linuxes anyone cares about. Similarly, you can understand why Red Hat and Novell want there to be only two players, because it simplifies their competitive rule books. And customers, who are short on time and money, just need two vendors of any product that they can grind against each other to drive the features up and the prices down. So Messman would seem to be right, if history and current attitudes are any guide.

But Linux is not a product. It is not something any one vendor has to pump a lot of money into to recreate to compete with the likes of Red Hat or Novell. Linux is an open source software stack that can be productized by any number of people and sold in extremely localized versions by people who only support local customers. This hasn't happened yet--but it could. Right now, the vast majority of entry and midrange server sales go through one reseller channel or another, and the reseller--not the server vendor--has the connection to the customer. (Dell still sells predominantly directly, of course.) Excepting their largest accounts and their biggest iron, IBM, HP, Sun, and Fujitsu-Siemens do not really own their customer accounts--they are suppliers to their channel partners. The Linux market could be--and maybe should be--highly distributed in this fashion. If the Linux Standard Base specification ensured application compatibility like it is supposed to, there is no reason to believe that I could not be using a local distribution of Linux created and maintained by a local team of techies who are right here when I need them. IT organizations might even like such hand-holding better compared to the take-it-or-leave-it approach of the Red Hat and Novell distributions.

The only barrier to entry for any group of people who want to deliver commercial support for any open source software component, and indeed, any commodity hardware product, is their own skills and desires. Instead of two Linuxes, there might be thousands. A company founded by a bunch of ex-Red Hatters called rPath certainly believes this is the future of Linux, and it has built the repository and Linux build system to allow companies and ISVs to build what it calls open source appliance stacks based on the Linux kernel. In effect, you pick and choose your stack, and then rPath keeps the pieces up to date based on what myriad open source projects are doing in terms of adding new features and security patches--and here's the important part--and you, as the customer, decide when you want to update your personal Linux stack and rPath does all the integration work to make sure all the dependencies between Linux components and other open source software such as compilers, middleware, and databases all work without conflict.

This is mass customization, and this is also something that most of the IT industry, which has been focused with laser-like intent on mass commoditization, is not exactly prepared to embrace. Or to defend against.

So maybe, in the long run, there will be a couple of Linuxes, including Red Hat and Novell but probably also at least a third and a fourth commercial distributor that offers software that is sold based on the traditional services model. And then again, maybe there will be zillions.