The Appliance Infrastructure Revolution
If we had to name a poster product to represent the entire embedded Linux world, it would probably have to be the Cobalt Qube. There was nothing quite like it before and nothing the same after. Here was an ISP-in-a-box that sold on looks alone. We had to put one on the cover of the October 1998 issue of our sister publication, Linux Journal. And, it has remained a founding example of a networked appliance though multiple generations.
Where Cobalt built most of its business, however, was with ISPs. Here blue Cobalt RaQs and other server appliances were the first web servers to be ``productized'' like consumer electronics, for sale to ISPs. As Cobalt CEO Stephen DeWitt explained to me a while back, the company wanted to make it as easy as possible for its ISP customers to sell bundled products to their customers, like any other appliances.
Cobalt also made a conscious choice to use the appliance label in a business environment where it had never been used before. In the process, Cobalt created the category. The result was racks and racks of blue Cobalt ``vending machines'' for network services (e-mail, web hosting, file services, security), database applications, cache, storage, streaming multimedia, ASP services, etc.--each bundled into a box and sold as an appliance to ISP customers. The concept attracted third-party developers, all writing applications to run on Linux inside Cobalt's boxes. It was a very successful concept, judging by the $2 billion it cost Sun Microsystems to buy Cobalt last year.
After the merger, Stephen DeWitt became vice president and general manager of Sun's Server Appliance Business Unit. We caught up with him on his way to an airplane in late January, eager to talk about two subjects where Cobalt has been a standout pioneer: appliances and embedded Linux.
ELJ: I've been looking back to see if anybody else was calling servers ``appliances'' before you guys. Would I be wrong to assume you created the category?
Stephen: You'd be spot on. Who else was there before us? Who went public before we did? The only other appliance company that would roll off anybody's tongue is Network Appliance.
ELJ: Another Linux company.
Stephen: Right. And they do cool things. There's also a lot of similar heritage there. We're the two that got the category started. They singularly focused on storage opportunities in the data center, whereas we're much broader in scope. In premise, appliances--what we did with the Qube--and definitely what we did in the hosting space, there was certainly nobody before us.
ELJ: How would you characterize that space today?
Stephen: The market between the router and the server is going to be characterized by appliance companies. We're delivering applications and content to everybody everywhere. If you want to call these the consumer appliances of the Internet that might be a little hyperbolic, but in terms of building appliance infrastructure for the mass markets, that's what we were birthed to do.
ELJ: What you're selling are ISPs that enables them to sell products to end customers.
Stephen: You got it. Bingo.
ELJ: I also think you rather cleverly conceived the idea of providing them with the inventory they can sell.
Stephen: Right. Now if you take the hosting industry as any kind of a benchmark of where we are and where we're going, just look back 24 months. It was like getting cable for a thousand dollars a month. Nobody would buy it and there wasn't much on. Now we're seeing the prices come way, way down.
ELJ: I was at the Appliance Pavilion at Comdex last fall, looking around at all these things called ``appliances'', which included handhelds and clients as well as servers. There was stuff from Network Appliance, IBM, Nokia, Motorola...I asked what operating systems each one ran, and most of them were using Linux. But a lot of them were not talking about it. My take-away was that Linux has turned into commodity infrastructure.
ELJ: So the question for me is: what's the Linux development conversation really about in the embedded market space we call appliances? Is it about the toolchain and specialized embedded Linux companies like Lineo, MontaVista and so forth--the Wind Rivers of this new space? Or is it about something else? Especially when there isn't a fight going on to attract attention? The victory of Linux in this space is a huge story as far as we're concerned. We wouldn't be in business if it weren't so. But in the absence of conflict not a lot of news is leaking out. What do you think is really going on here? Why should somebody roll their own Linux vs. using Lineo's or Red Hat's?
Stephen: It's funny, because we started out with Red Hat and a little customization. Now we're four years and five generations down the road, but we still start with the basic Red Hat distribution because when you court developers--whether for embedded or mainstream applications--you need a common reference point. That's what using Red Hat gives us. But it's in the middleware that we've invested our value add. We're 100-plus engineers for a number of years' worth of work, building a platform on top of the operating system. You are absolutely right that the embedded market was won overnight by Linux and nobody knew it. What we've all gotten savvy to is that nobody wants to play at the operating system level. And, nobody knows if there is a business model there. Is there a sustainable business model there for Be? I have no interest in that business. What we have done is put all of our management technology, all of our clustering technology--things like active server page integration on non-NT platforms, linkage for third-party developers to build their application and link into our UI or our other toolsets quickly--all the classic middleware functions, on Linux. That's where we put a lot of our energy, so anybody who is building any kind of application on Linux can very quickly leverage our appliance value proposition and deliver an appliance to market. And that's very compelling for certain classes of applications. So we've got close to 3,500 developer partners with just a four-year-old company. Everybody from Bob's software to Main Street database, streaming, caching, e-commerce, storage and a whole slew of verticals--medical, legal--all being delivered as appliances.
Stephen: Yeah. I've got to deliver a guy this box that he uses as an application. He pushes an ON button starts using what he paid for. He doesn't dicker with CD-ROMs or operating systems.
ELJ: What about the development side of things? How is developing embedded Linux different what came before?
Stephen: One of the problems in the traditional embedded space was that you had kind of proprietary ways of doing things. That gave companies like Wind River a lot of control. In the Linux world, you have a referenceable, all-world operating system that everybody is exposed to. You're not married to any particular distribution. You can go off on any tangent and add the value you need for any particular application. It makes a hell of a lot of sense to me.
ELJ: It's two-by-fours.
Stephen: Exactly! A good way of looking at it. You're not going to Joe's lumber company and having him sucker you into something that works just the way he wants to do it.
ELJ: Would you say your 3,500 developers are developing for Linux?
ELJ: So, sitting on top of the Net, Linux is additional commodity infrastructure?
Stephen: Yup. Good way of looking at it.
ELJ: So it's a general-purpose operating system that lends itself to all kinds of single purposes. Then you turn those into various appliances.
ELJ: Let's look at what you call the ``premise'' space. This is where you introduced the Qube several years ago. What's happening there now? What's the future of the Qube?
Stephen: It's just blowing me away. It's a product that, from the day it was first featured on the cover of Linux Journal, all the way through its heritage, has just been doing great. It's an ideal premise appliance for any big service provider looking to reach down to the other end of the wire. You're going to see us do so much more here. We've announced two major deals so far: one with NTT/DoCoMo to create a wireless premise appliance and one with France Telecom to create a dedicated educational appliance for the French market. These are big service providers who reach millions and millions of subscribers and who want to create a device where they're delivering some sort of network value-add at the customer's premise. We are in the works on a bunch of other things, but watch very closely at what we're doing this next quarter, where either big cable providers or telcos are going to hang this sort of class of consumer appliances at the end of a leased line, or the end of a DSL line or some other kind of line, so they can serve up some content or applications at the customer premise. Those could be very far-reaching. It can be everything from home networking applications to home web applications, to entertainment applications. It's kind of like having a jukebox at the end of the wire.
ELJ: In the same way as you introduced the appliance concept to what we might call a collocation of services, are you going to branch out from there? Or are you going to concentrate on continuing to define what happens where you're active today?
Stephen: First, we have a lot of products on the drawing board right now. You're definitely going to see us in the enterprise where our class of infrastructure has never been before. And you're going to see us down at the other end of the wire, which not only is a new business for Cobalt but a completely new business for Sun. I mean, Sun has dominated the big enterprise, service provider and telco markets, and appliances are a new opportunity for them, both down-market and up-market--also new products. We're going to deliver a whole new family, not just premise and hosting appliances.
ELJ: How would you define an appliance?
Stephen: Fundamentally, it's a dedicated piece of infrastructure that is optimized to deliver a single application or a set of applications. It's not a general-purpose platform designed to run anything. It's optimized to deliver a networked application.
ELJ: So single purpose rather than general purpose?
Stephen: Well...single purpose isn't right. We have multipurpose appliances, but they are not general-purpose computers. We have hosting appliances delivering mail and web and network services. It's not single purpose but rather optimized for a hosted customer. It's tailored for delivering networked applications, first and foremost. It's not client/server. It's about network applications. And there is no exposure to an operating system. There are no CD-ROMs, no monitors, no keyboards. The total cost-of-ownership model around an appliance is measured in dollars, not thousands of dollars. Those are critical components of the story as well. You never need to go down to a command line. We're talking about consumer-class ease of use and reliability. Meaning you're more likely to reach mean time between failure with the drive than the application.
ELJ: That is an interesting characterization.
Stephen: Think about it, when was the last time your microwave died? We want five-nines reliability (99.999%). Traditionally you can't get there without spending hundreds of thousands of dollars. But when you build a stereo, you expect those five nines. When we first sat down to build these appliances we asked, ``How can we build a sub-$1,000 box that anybody can use with any level of expertise, that will never fail, for any particular application?'' And, multiple generations later, we've pretty much perfected that model.
ELJ: The model of reliability.
Stephen: Also consumer-class ease of use, reliability and pricing for networked applications.
ELJ: What about web client appliances?
Stephen: Not interested.
ELJ: I have a theory about those. When you're sitting at your desk looking at the Web, you want a general-purpose device. You want a personal operating system.
Stephen: For us it's just a receptacle--a socket.
ELJ: So what you care about is boxing up services and selling them to providers in ways that make it easy for them to sell to their customers, who have those sockets out there.
Stephen: Yeah, at the end of the day, when you think about the evolution of the cable industry, and that little box sitting on your TV, you don't care about what operating system it runs. You pay your two bucks, and your content is there. Somewhere back inside your service provider is all the infrastructure that makes that happen. And it always works it's always cheap and incredibly easy to use.
ELJ: It's interesting to me that those five nines are now the most basic value statement.
Stephen: If you're going to sell to the Fortune One Billion, you can't sell what fails. The Fortune One Billion can't deal with failure. We wouldn't have 98% of homes with television if TVs were hard to use, or if every now and then we had to reboot one.
ELJ: Will you ever really get to those five nines with general-purpose personal devices, with PCs? Aren't they just a little too open-ended in purpose?
Stephen: Yes, to some degree. As soon as you open the door you introduce things that can hurt reliability. If you create a solid state, there's overhead involved with that as well. The fundamental thing that changed everything, and what made Linux so hot in the first place, is putting developers into a billion-person marketplace. Before, with PCs, it was one person. As soon as you make the billion-person context your backdrop, you change the requirements for the way you architect and deliver applications. It's changed forever. That's why the role of an operating system, which a few years ago dictated evolution in this industry, is infrastructural. The old way went with the dodo as soon as the Internet showed up.
ELJ: I think computing got social when it started to happen on the Internet. It was no less personal, but it had a social context. And a lot of functionality will sit out there in the social context.
Stephen: You got it.
ELJ: So you've located a lot of your reliability on the server. The strategy is to locate as much reliability as possible on the server as a way of dealing with the inherent unreliability that shows up around the client.
Stephen: Absolutely. You have to take out the complexity--put it back in the server. That's one thing. The other is to create a class of infrastructure that really does the job. If you have to cut down a tree, would you rather do it with a pocket knife or a chain saw?
ELJ: And the applications follow.
Stephen: The microwave is a good example. It may have the general purpose of cooking food, but it's simply best at cooking certain kind of things, like bags of popcorn. So now you see all kinds of people creating food for cooking in microwaves, just like people are creating applications that can be serviced in an appliance.
ELJ: Let's go back to Linux again, and what it's doing to the embedded world. I have a theory that what's going to happen with manufacturing will be similar to what happened with publishing a few years ago with the PC, or with starting up a business that uses the Web. The threshold of enterprise is getting lower and lower as the cheap and easy infrastructure starts to appear. I'm wondering about big manufacturing businesses that specialize in small things--such as consumer electronics. It seems to me that it's getting easier and easier for anybody who's motivated to get into the appliance business and to diversify that category. If I want to put a TCP/IP stack and some intelligence on my stereo, I can do that. I can make the whole stereo, or a component, whatever. There are Linux geeks out there making MP3 players for cars--already.
Stephen: Absolutely. It's a new game. Take a look at our lab some day. Everything you can imagine is being thought of back there, and we're working on it, lighting systems, wireless you-name-it. I don't want to call this the Radio Shack crowd of the next millennium, but the fact that you can now whip together all sorts of stuff is really significant. Some of it is going to stick. The good thing is that the rate of change is getting even more rapid.
ELJ: And the cost of failure is getting lower.
ELJ: So you can experiment a lot more.
Stephen: You bet.
ELJ: And we'll be seeing a lot more stuff nobody expects.
Stephen: And a lot of it is going to be crap. And some of it will stick.
ELJ: And even manufacturing, which used to be something only big suppliers did, will get more diversified, small scale, adaptable.
Stephen: Yup. You got it.
ELJ: There will be guys who make plastics for less money. Others who make boards for less money. Others who integrate software on chips for less money.
Stephen: You got it. Those are the economics of internetworking. It's because we've gone up to a size of potential market that's so much bigger. The costs are also coming down accordingly. Economics 101 is being retaught in the age of the Internet.
Doc Searls is senior editor of Embedded Linux Journal and a coauthor of The Cluetrain Manifesto.