CRN Interview: Linus Torvalds

CRN: Linux 2.4 is now out the door. There are thousands--if not millions--of open-source developers contributing to the Linux kernel, but ultimately, you alone must sign off on it, correct?

Torvalds: Yes. In the end, it's not that horribly hard. The worst part is just the herding of developers and trying to cajole them into doing the right thing--or at least make sure they don't get too upset with me when I say no to something. Another horrible thing is having the press trying to second-guess me on when the release will happen.

CRN: As a man with a full-time job at Transmeta, is it becoming too much for you, in terms of energy and what you can devote to development of the Linux kernel?

Torvalds: I'm lucky, or maybe I have just made the right decisions in life. Transmeta has been very supportive: When I said I needed to concentrate on getting [Linux version] 2.4.x out, the people here basically asked me what Transmeta could do to make it easier. And it does help that I've been doing this for 10 years. Making a release is fairly stressful, but at the same time, it's something I've done before. And I have a fairly good feel for it, which makes it much less stressful.

CRN: But the Linux 2.4 kernel was more than a year late. I remember that it was scheduled to be released by year-end 1999.

Torvalds: The original plan was to try to aim for a nine-month productization cycle, and part of the reason for that was that I expected the changes for 2.4.x to be much smaller than they ended up being. My original main goal was to clean up the SMP scalability to four CPUs, and it kind of grew into a major file-system redesign. That said, everybody knew the nine-month goal was unrealistic. But it was kind of "if we don't have anything to shoot for, we certainly won't hit it." I was hoping we could get it down to a year or so.

The gap between [Linux version] 2.0 and 2.2 was well over two years and had been quite painful. [The update to 2.4] ended up being almost two years, and while it never became as painful as the 2.0 to 2.2 gap--probably because 2.2 was maintained more actively--it's certainly still true that two years is a bit too long.

The good news is that for the 2.5.x series, we don't have anything like the 2.3.x wish list. Of course, that may change. I'm really happy with how 2.4.x ended up, though there's the obvious few months of watching stability, etc. Maybe the 2.5.x series will be less intrusive, which would translate into a simpler release next time around. Knock wood.

CRN: If there was a Linux board, such as the Linux Lab, with 20 engineers devoted full-time to finishing the code done, finalizing it and releasing it as soon as possible, do you think the same delay would have occurred?

Torvalds: Probably. Delays are kind of inevitable in this business. Of course, because I never really had any hard deadline that I set for myself--my only criterion was really "when I'm happy with it"--I probably didn't get as hung up about the release as an "official" body would have, and maybe a Linux Board would have held people to stricter deadlines. Who knows? At the same time, I think 2.4.x happened when it was ready--not before, and not later.

What didn't get very much press coverage was the fact that we really did have a Linux Lab working on testing. In fact, there was more than one. All the major vendors had a 2.4.0 stress-burn setup, and we found quite a few bugs that way. I'm really happy with the level of support that I got from Linux vendors in this area.

CRN: Windows 2000 was years late. But that was Microsoft's woe, and its channel was upset, too. In this case, your partners--Red Hat and other Linux distributors--had their products ready to go in the fourth quarter of 2000. Weren't they upset when Linux 2.4's release was delayed?

Torvalds: So far, the vendor comments I've gotten have been pretty philosophical. Many of them expected it to be even later, actually, and I never got a complaint. Most of them seem to have covered the kernel development fairly well, and I think they'd have been as unhappy as I would have been if I had released 2.4.0 too early.

CRN: How much pressure did you get from Linux companies, whose profits were hanging in the balance until 2.4 shipped?

Torvalds: None at all, really. To some degree, I got "what do you think the timing will be?" kinds of questions. But absolutely no pressure. Nobody said "we'd really like it for the Christmas season" or anything like that.

CRN: But how about the kind of pressure that IBM, Intel and Oracle can exert when they're hungry to ship product? Are you feeling more internal pressure, given the vast constituencies in waiting?

Torvalds: Forget about the Oracles of the world. Their product time lines are so long, and they want to test things on the shipping version. So they really don't have any reason to even ask me. Only now that 2.4.0 is out will they even start to think about it becoming an issue. Intel obviously has the IA-64 issue, but at the stage they are in shipping the hardware, they were happy with even a non-released 2.3.x test kernel. Having a real stable IA-64 2.4.x kernel out probably won't be an issue for them until later this year.

CRN: You alluded to a time in the future when you won't be the one in charge of Linux. Who or what will be? Is there a new company, entity or organization being prepped to take the reins from you as guardian of the kernel?

Torvalds: What will be, will be. You ask the wrong questions.

CRN: If a large, full-time team of engineers directed by a full-time leader worked on the Linux kernel for the past year, might some of the features planned for 3.0 made it into the 2.4 kernel?

Torvalds: Ah, the old question about discipline as the best way to reach your goals. Is an army known for it's thoughtful approach? Would Mozart have been more productive if he had scribes to help him, a secretary and a CEO to lead his way? What if? We'll never know.

However, we do know that the sometimes disorganized approach--the strange and motley band that defied traditional ways of writing software, the crowd of individuals rather than the sternly disciplinarian army of ants--has so far done a lot better than anybody expected. So far, we have a track record of having done a lot better than a lot of "disciplined" companies, with full-time leaders.

CRN: Do you envision a day when you won't be the one signing off on the final Linux kernel code?

Torvalds: Some day, sure. Not in the near future, though. I don't plan that far ahead.

CRN: Might you hand over the reins to the Linux Lab in the next two years and serve as an adviser rather than lead developer? Might you sell the Linus Torvalds trademark to a company, entity or other person? Might you license it?

Torvalds: Ah, you have discerned my plan! I will have to kill you before you disseminate it to the rest of the world.

Seriously, I don't plan two years in advance. Cross my heart and hope to die. I can be pretty sure that I won't "license my name" or anything like that. If the day comes when somebody else does a better job of maintaining the kernel than I do, I would hate for him to work under my name. The guy who gets the job, gets the glory--and the blame. And, yes, it probably would be "the guy" or "the group of people who trust each other" [to maintain the Linux kernel] and not a company or a "Linux Lab." I'm not saying there is no space for companies, rules or regulations in Linux. They obviously do their thing, and it just so happens that "their thing" is not something I get all that involved in. As a technical guy, I have other issues.

Maybe I'll be around when I'm 70. I'll try to age gracefully, turning into a gracious older gentleman who still holds the reins. That would be kind of scary, though--the notion that some young, eager whippersnapper hasn't come along and done a better job than me by then. I'd consider that scenario unlikely.

CRN: What features are planned for 3.0? Will the development process be different, and have you selected any one to work with you?

Torvalds: There are a lot of issues we'll work on. We pretty much fixed the file system and [memory management] for the time being, but better generic support for things like journaling file systems is still a thing to be done. Maybe for 3.0. Right now, the different journaling file systems all have to provide their own journaling capabilities, with very little commonality. That's fine at this stage and probably will be for a few years to come, but we have to plan for when it isn't the case any more and be ready.

Clustering [also will be a focus for 3.0]. NUMA architecture support. Big iron. Parts of it are there, others are coming. But let's face it, there will be a lot of companies and people wanting to work on sexy, big hardware. But there will also be a lot of people working on the fun, small hardware. Much of that is much more experimental and less directed, but because of that, also more fun and inventive. Big iron has been done before, and we'll do it better. But one of the things that makes Linux stand out already is the width of support. I think we'll see tons of interesting stuff happening in the small embedded space, but because there is less precedent, it's much harder to point fingers to it.

And no, I don't select the people I work with. People tend to select themselves.

CRN: The Linux movement has begun to fracture, with companies such as TurboLinux modifying the kernel. Under this scenario, how can you control Linux's future development?

Torvalds: Since when is it news that people and companies involved with Linux do their own thing? That's the whole point with it. That's how things get done. That's what open source was and is all about. The good ideas will flourish and get incorporated into other versions, and eventually back into the "standard kernel." The bad ones were often still worth at least trying out. It's an electronic jungle out there, and evolution is the name of the game. Some species work really well without evolving for long times--the sharks, frogs and tortoises of the software world. Others evolve to become something more.

CRN: Microsoft CEO Steve Ballmer has finally said Linux is the No. 1 threat to Windows. What's your response to that?

Torvalds: "Tag, you're it." I don't care. They've had a lot of enemies in their time. Let them fight one enemy that doesn't care for a change.

CRN: Sources have hinted that Microsoft's Common Language Runtime, in Visual Studio.Net, will natively support Linux, Solaris and IBM S/390 in the near future. Is this your understanding?

Torvalds: I have no idea. I stopped following Microsoft rumors a long time ago. I'd be surprised if they were willing to start loosening up their stronghold on the platform, though. They've come a long way by leveraging the basics. I doubt they'd suddenly change their mind.

CRN: Do you think Linux 2.4 is ready to take on Windows and/or Solaris in the enterprise space?

Torvalds: "The enterprise space"--why do people pay so much homage to that thing? Face it, what makes everybody respect Microsoft as a force in the marketplace isn't the enterprise space. They just ignored it and ignored all the analysts that said DOS and PCs are just a joke. Go back 15 years and think about who the masters of the enterprise were back then.

Linux is ready to tackle the enterprise space for the same reason Windows NT was: because it has grown up and encircled it. And the enterprise space will do what it always has done: notice that it was overtaken by the computers and systems that it really honestly thought were just for hobbyists and people who couldn't afford "real hardware." In technology, the only thing that counts is volume and market share. Linux happened on both, simply because of how it is developed and what I think is a better way to compete. It's all about competition, and people forget that competition is not about who has the most money to spend.

CRN: Is there an inherent conflict of interest for a processor company--Transmeta--that profits from Linux, given its free, open-source nature? How much have you earned from your Linux [trademark] and development efforts?

Torvalds: I can honestly say that I've made more money from things that people chose to give me freely than I've ever made in my entire life from "honest work," Transmeta included. And you know what? That actually makes me damn proud. And it doesn't make me belittle my "honest work" at Transmeta in the least. They are two parts of what I do. I'm not a black-and-white person. I like them both. Are there conflicts of interest? Hey, anything in life is a conflict of interest to some degree. You are what you make of them.

CRN: Ultimately, customers want a corporate entity they can hold responsible for their technology investments. Might you, along with some of your top Linux developers, form your own company?

Torvalds: I'd rather keep my fingers away from forming a company, thank you very much. I don't need that kind of headache. This is what commercial Linux and non-Linux companies do everyday already. They act as the corporate face of Linux. And that's fine. I certainly never wanted that part. So let the people who want to worry about that side do what they do best. I'll just continue to plod along at what I do best--or, at the very least, at what I find most interesting.

CRN: What is the least understood aspect of Linux? What don't we "get" when it comes to Linux?

Torvalds: I think most people get it now--not the whole picture, necessarily. But then, who am I to say that my whole picture is the right one? The important thing is people do realize that great technology doesn't come from great companies. Technology comes from the people involved, and the companies are really only a corporate facade for that. And that corporate face is not always necessary, and it is not necessarily always a good thing.