Online Extra: Q&A: Can N1 Make Sun No. 1?
The future of Sun Microsystems (SUNW ) hinges on a technology vision dubbed N1, for Network One. Sun engineers want to use N1 to make computer networks more reliable and behave almost like an electric utility. Greg Papadopoulos, Sun's chief technology officer, recently spoke with BusinessWeek Silicon Valley Correspondent Jim Kerstetter about Sun's vision for utility computing. Edited excerpts of their conversation follow:
Q: What computing problem is Sun trying to solve with its N1 strategy?
A: Of your readers' information-technology budgets, about 80% of them go to paying people for running IT. All these things like Linux and Intel -- they're just 10% of the budget. Other software, the applications, are maybe the other 10%.
Q: But it's not so easy to cut from that 80% operational expense. Isn't it a lot easier to cut the other 20%?
A: Partially. I can take two systems that are sitting in a network, just 15% utilized, and combine them. Right there I've already cut my operational costs in half. But that's a short-term thing. That doesn't answer the fundamental question of "Why is this stuff broken in computing?"
Q: So you really believe this is the next era of computing?
A: We think a lot of pressure is building up for this new era in computing systems. This new system -- instead of being driven bottom-up from new technology -- is being driven by a higher-level view of computing.... Developers are [now] thinking about writing a new service for a network.... They're worried about getting tens or hundreds or thousands of machines to work together. It's a much, much richer environment.
Q: So how then do developers view N1 and your Sun One suite of infrastructure software?
A:If I look at this stack of storage and networking and servers and operating systems, this middleware we're talking about, the applications on top -- Sun One is the developer's viewpoint of this stack. It's the platform for Java [programming language].
In an interesting way the industry has condensed around just one or two views of this -- Java and Microsoft. Many companies had the opportunity to take advantage of this, the HPs and what have you, which did not.
Q: That's a fair point. But didn't your rival IBM "get" Java from the get-go?
A: They did understand the importance of Java. But IBM is getting into a very interesting conflicting situation. If IBM truly goes after the systems engineering that tries to go after the 80%, that goes right after all the outsourcing and services that IBM has. They have more of their company built around assuming the complexity for people.
When you're in that position and you really do something cathartic to solve that complexity, does that go after your own business in a big way? I don't know. If you look...at IBM, they have suppressed innovation inside the company because of the threat to their franchise.
Q: But isn't Sun's franchise the Solaris operating system for servers? Won't sales of that be impacted by this change?
A: Not really. We're finally breaking out and saying that the operating system is no longer a thing that controls something within a box. It's something that can control a set of network elements. We've never really built something like that in computing. We've had things like Novell's network operating system. But that was a misnomer. As an operator, I don't want to look at a thousand different systems. I want to look at an operating system that collectively handles all these resources.
Q: So then how long does it take to get to that N1 vision?
A:You could ask that question how long did it take to get the PC vision -- or the minicomputer vision? These things all have a classic adoption curve. They all last in terms of their peak and contribution, from a margin point of view, to a company.
It's a decade-long phenomenon. And then they become increasingly off-the-shelf, and then the business starts going to efficient producers and things. I think we're at the beginning of this era. There are certainly things you can be doing today, and we're expressing today what can be done in a short period of time, a matter of this quarter or next quarter or the next year or two years.
But the hard work is just starting now. I won't say it's going to be this percentage of revenue. We'll begin to shake that out over time.
Q: You're pretty confident. Why?
A: It's like Sun's shift from workstations to servers [in the early '90s.]. There's a very important lesson in that. A lot of our early servers were workstations that we removed the consoles from. And we moved Solaris forward to become more a server-style operating system. That was a lot of that early business. Similarly, a lot of the early N1 business right now is taking a lot of the servers we have now, teaching them how to be better participants in a network, without having done anything deliberate about the hardware underneath it.
So in a very real way, it's almost certain that the way we think about processors, the way we think about memory, storage, and networking, are wrong. When you get down to the fundamentals, maybe I shouldn't be building the biggest bad-ass processors I can make.
Q: You're still investing heavily in chip development, I guess, to get closer to a type of computer chip that has less horsepower but can do things like improve software throughput. How close are you to completion?
A: We're very far along on that. We've made a lot of investment over time. I won't tell you how close we are to product. But it is a major point of recognition for us that we believe what's taking place in processing is a true bifurcation. We have one class we call internally the "boomer class," named after the big nuclear submarines. And only three companies in the world are doing that anymore. Sun. Intel. And there's IBM.
And there's chip multiprocessing, where you put lots of smaller processors together. You evolve that over time. The first category is about capability. The second is about what kind of workload can I put together? It turns out that throughput processing is a really powerful concept in this. People don't build Web-server farms out of big, bad systems. They take a lot of smaller systems and rack and stack them, and perform load-balancing with them. It's a form of clustering. That's where throughput comes in.
Q: So how does all of this get to what Sun sees as the computer system of the future?
A: The new system is really pretty easy to state: It's the computer that you build out of the network. That's it. If the network is the computer, then the computer is the network.
Q: What's inside this network computer?
A: All the things that we used to call systems. They were the servers and storage and all that. We used to build servers as stand-alone computers. Now we're using those servers as elements of this network computer.