sign in sign up
<<
Home , Linus Torvalds

News Interview: Torvalds

Interview with Courtesy of Foundation

In June of this year, kernel developer Greg notion that we need to have a feature to in- five to eight years, when we decided we’ll Kroah-Hartman sat down with Linus Tor- crement the version number – and then we be stable all the time. I think it’s helped us valds in front of a live audience at LinuxCon changed how we do development. as developers, but I think users have been Japan to look at the first 20 years of Linux, much happier, too, when they don’t need to We don’t do versioning by features any- the state of the kernel, and the future of worry too much about being on a new ker- more; we just do this 8- to 10-week release Linux. nel version. cycle – and it’s working wonderfully – but Linux 3.0 it’s meant that 2.6 has stuck around for a Growing Up Greg Kroah-Hartman: You announced that long time. Now, 20 years of Linux means GKH: In the 20 years, has there been any re- the version number is going to change. You that I finally had the excuse to say, “OK, cent features you liked or found interesting? said 3.0. enough. We’ve done version numbering LT: I actually like the really boring features. Linus Torvalds: It’s not out yet, but I did the based on time, so let’s change 2.6 based on We’ve had many performance improve- RC just before I left for this trip. If everything time, too.” Now it will be 3.0, and you [indi- ments in the last few releases. The one I like goes well – and it looks fine so far – in about cates Greg] can do 3.0.1 for your stable. particularly is the name lookup changes. It seven or eight weeks, we’ll have the final We’ll still have a lot of numbers, but the speeded up enormously on some of my 3.0 release just in time for the year’s festivi- numbers will be smaller, and I don’t think loads that I run. We literally had a 40 per- ties. I’m actually really happy about the we’ll ever hear 40 anymore. When we get to cent performance improvement in one re- whole thing. I’m finally getting rid of 2.6! 3.20 or something, then I’ll just say, “Hey! lease, and that’s basically unheard of. But Let’s increment to 4.” there’s no new feature. There’s no new in- I’m finally getting When I made my 3.0 release candidate, I terface for users. There’s nothing new going created a diff to the previous version. I cre- on, we’re just doing it 40 percent faster. rid of 2.6! ated the diff not against 39 but against 29 – That, to me, is really exciting. because when you’re talking about big GKH: Doing things faster is a nice change. I don’t know how many of you know this, numbers like that, they all look the same. I Normally, the joke is that we’re getting but with our old versioning, I added a third didn’t notice I was starting to upload this bloated. number – we’re up to 39 right now – and diff against the wrong kernel version. I’m then to make things even more interesting, hoping now that we’ve renumbered, it will LT: Well, we still are pretty big. Thinking Greg does something like 2.6.39.1 to num- be easier to remember. about the machine that I ran Linux on 20 ber the stable releases. It just gets really years ago, we wouldn’t fit on that machine GKH: Thank you. I deal with these numbers messy. Then all the distributions have their anymore. The good news is that even a cell- every day, and it drives me crazy. own build version, so when you actually phone has 10 times the computing power run Linux, you will run something like LT: I hope this will clarify things, although, of that machine these days. Nobody really 2.6.39.7-13 [audience laughs]. We’ve been right now, it’s also causing a lot of discus- cares that we do need more resources; we doing this for a long time, and it’s been kind sion about – do we start doing new features use those resources very efficiently. We of meaningless. now because we’ve changed the version? have been growing, but I think most of the No, it’s all the same, we just changed the We used to change the version number growth has been to do things that modern numbers so they’re easier to remember. when we hit a big milestone. So, 1.0 was hardware needs to do, things that modern usage patterns need to do. networking works; 1.2 was our first multi- GKH: People have said, “Can we remove platform, so that was when we supported things?” People want to get rid of micro- GKH: I know a lot of kernel developers have alpha, m68k; 2.0 was SMP worked … kind channel or ISA or IDE. objected to the interface, the con- of … ; and 2.2 was SMP actually scaled to trol groups, and some people have argued LT: No, we’re not getting rid of features. One two or three CPUs. We’ve always had this that it didn’t have initial users, but now we of the things that has worked so well for the do. last couple of years is trying to be stable all The Video of the time. We used to have these big LT: The cgroups interfaces were mainly The original video was produced by The jumps when we removed features and re- done for certain server setups. Not very at LinuxCon Japan. To wrote the kernel and did “big” develop- many people really used them. A lot of peo- see the entire video, go to http://www.​­ ​ ment changes. It was necessary back in the ple were unhappy because they compli- ­.com/​­ watch?​­ v=__fALdvvcM0​­ . 1.0 days, back in the 2.0 days. You had to re- cated the memory management code, they write the world. It’s been so easy in the last made the scheduler more complex, they

16 October 2011 Issue 131 linux-magazine.com | Linuxpromagazine.com

016-018_Linus.indd 16 8/10/11 3:39:30 PM News Interview: Linus Torvalds

had a lot of impact on very core infrastruc- people thinking of one particular platform LT: Some people flitter from project to proj- ture. Then, it turns out, we ended up finding and not thinking of themselves as being a ect, and I’ve always been somebody who great use for it outside of the original target part of a bigger ecosystem. ARM is growing likes concentrating on one thing. I’m not a audience. It’s interesting how often it does up; along with that, they’re starting to hit all multitasker. I’ve had a few small projects in end up happening. Like all the SMP code. the same issues that the other platforms hit between, but I’m really happy doing one Ten, 15 years ago this was the “big iron,” a long time ago. thing and feeling like I’m doing a good job the server feature, and obviously today, at that one thing. I didn’t think I’d do it for 20 every single high-end cellphone ends up Linux Everywhere years, but now, if I didn’t do it, I’d be bored. using SMP code. GKH: Speaking of ARM. Linux now is every- GKH: Do you see yourself doing it for an- where. It’s in our small cellphones to … ev- That’s one of the strengths, I think, of Linux other 20? erything. A long time ago, when you did the because we have the same kernel across so 1.0 release, you said “total world domina- LT: I’ll be old by then [laughs]. I mean, I was many platforms. Nobody else does this. tion” was a goal, and we’re doing pretty really young when I started, and another 20 Apple has iOS for their low end and OS X well. But one of the things you said then years … at some point there will be some- for their high end. Microsoft has Windows was for us to achieve that, we’d have to body young and hungry and energetic who CE for low end and “real” Windows for high have the applications. The kernel was bor- comes along and shows that he’s really end. Linux has never had that. I think it’s good at it. That’s the point where I think I’d one reason we are doing really well in the ing, and all the interesting stuff was going stop. I’m perfectly happy to say, “Hey, embedded space. We never had a cut- to happen in the applications. Do you think you’re clearly doing a better job than I. Go. down, kind of castrated version for the low that’s still true? Take it.” end. We always had full features because, LT: It’s less true than it used to be. We have as it turns out, all those high-end features GKH: How do we keep Linux relevant to be the applications to a large degree. I don’t eventually end up percolating down. able to hand it off in 20 years? How do we say “world domination” anymore. It was make sure we succeed? GKH: In the low end, the embedded guys funny 15 years ago because it was so obvi- argue “nobody’s going to care about power ously a joke. For the last 10 years, it’s not LT: I really don’t think that’s a problem. If management but us, so let’s do it in our been so much a joke anymore, so it’s no you look at all the work we do today, a lot of own little tree.” But [look at] Google’s serv- longer funny, so I stopped saying it. it is hardware maintenance, and that ers – people have real power needs and big doesn’t seem to be going away. A lot of iron. what the kernel does is day-to-day stuff I don’t say ‘world people did in the ‘60s. LT: It goes that way, too. Power manage- ment grew up instead of growing down. domination’ anymore. The whole architecture is 40 years old, and I don’t think it’s any less relevant today Playing Nice We’re doing really well on the low end; than it was back then. I don’t think 20 years GKH: That leads to [the idea that] people we’re doing really well on the high end, will make a big difference, but we will have sometimes feel their features are just for servers – pretty much every little niche. The to update for new hardware and new usage them. The ARM community has been very, desktop is where we actually have the ap- cases. With new hardware comes new soft- um – how do I put it nicely – insular, think- plications now. It’s a hard market to get into, ware and new places where that new hard- ing that they’re doing things only in their and it’s still the market that I started Linux ware gets used. So, I think we’ll be relevant own sandbox, so it’s not going to affect ev- for. I wanted it on my desktop. It’s what I in 20 years. erybody else. For years, we’ve been trying use every day. I have a Linux phone and I’m to push back, and recently, you pushed The one thing I don’t want to be is in main- really happy about that, but it’s still the lap- back really hard. Do you want to talk about tenance mode, where we don’t live, where top and the desktop that I really work with. how that’s working out? we don’t make changes. It just takes a long time to convince people LT: I’m actually happy to talk about how it’s GKH: Yesterday at the open forum discus- to change what they’re using, so they’re still working out. In 3.0rc1, one of the things I’m sion here, there was lots of talk about how stuck on Windows and some are still stuck really happy about was the ARM people software can help with disaster on OS X, and we’ll get there some day. were starting to react. A couple of months relief efforts. Is there any role do you think ago, we had a big flame war, and I basically GKH: Can we do anything to help that in the that Linux can play in that? called them bad names, and people were kernel, or is it just working with people? not very happy, and some people thought it People are using was just me being difficult – which some- LT: I’ve been trying to think what we can do times happens. in the kernel, and I don’t know. We have Linux for things I worked very hard at making the kernel do The ARM people are realizing that they as well as it can. We have worked hard on never envisioned. shouldn’t think of only their small platform the whole interactive performance and user LT: One of the things that personally is very but try to make the code generic so they interface issues. There are still application gratifying to me is how people are using share code among themselves, which is a issues, but I think it’s really up to distribu- Linux for things I never envisioned. Not just big step. Now, they’re also trying to share it tions to very aggressively target the desk- outside of ARM. We’ll see how well that ac- the markets, but people using Linux and top. I’m happy that some of them clearly tually works. It at least means that now the open source for reasons that were not my are. people outside of ARM that are working on reasons – going into developing worlds and similar features see the ARM code, because Past and Future spreading knowledge of technology and it’s not hidden somewhere in the deep con- GKH: So, 20 years has been a long time to making it a teaching tool. I find that to be re- fines of the ARM tree. There’s a history of work on one project. ally exciting and gratifying.

linux-magazine.com | Linuxpromagazine.com Issue 131 October 2011 17

016-018_Linus.indd 17 8/10/11 3:39:32 PM News Interview: Linus Torvalds

Disaster relief – I’m sure it works really well. replacing a lot of local applications, but I which has always irritated me immensely I must admit, it’s not something I’ve been think we’ll still have a lot of native applica- and has sometimes been a problem for personally thinking about. tions, and I don’t think, for example, that the users, too. It’s kind of been going away Unix background of Linux is ever going to since Linux has been growing so much. A Questions from the Audience be a problem or something we want to for- lot of the manufacturers are finding out that Q: What is your single most memorable get about. The Unix architecture has been it really hurts them not to help us. moment? very successful. One of our strengths is that LT: That really is hard to answer. It’s been we were able to build on top of good design small ideas with lots of different people made by smart people long ago. The toughest being involved. It’s plodding day-to-day Q: How happy are you with ? And if problems have never work, and when you look at it a year or two you’re not happy, what would you change? or five years afterward, you say, “Wow, we been technical. really made a huge change,” but at no point LT: [Sighs] This question should actually go was there the “Aha!” moment. to Greg, because I suspect the answer from The other big problem is that it’s hard to de- Greg would be way more amusing because The one single moment for me, personally, velop a big project with thousands of peo- he feels much stronger about this than I do. was back 19 years ago, when it went liter- ple and tens or hundreds of companies that I think Ubuntu is interesting because they’re ally from a personal project to being some- are major contributors involved, and they taking a different approach from a lot of thing where I no longer knew the people in- have completely different ideas of where other Linux distributions. I don’t mind that. I volved. Now, it was not my toy anymore. they want to go. So, there’s been many think it’s been very helpful to have a distri- Now there were hundreds – at that time it times during the 20 years where we’ve had bution that takes a very different and maybe was hundreds – of people using this project big disagreements between developers. less technical approach and a more user-in- We’ve had people who were really unhappy of mine that I never met. That was a big terface and user-centric approach. Ubuntu, I with how development was done. We’ve thing for me. think, has been very successful, thanks to had people who were really unhappy about Other than that, there’s been lots of exciting that. That’s good, and it’s showed the other their feature not getting used when some- developments, like when Oracle announced distributions to some degree a piece that body else’s feature was picked. If I lose they’d port their database to Linux. Now they were missing. At the same time, we’ve sleep over something, it’s always about pol- we’re in the big league because if you are a had some issues at the kernel front, where itics and people. I’m happy to say that I Unix and you have Oracle running on top of some kernel developers who I won’t name think we usually solve our problems, but you, you’re a real Unix. But for me, person- [pointing at Greg] felt that Ubuntu wasn’t we’ve had times when we’ve had really bad ally, it was when it went from this pet proj- pulling their weight and helping as much as blood in the community. ect that I taught to a few people that I knew they should. Q: Are you still happy with the license, or do to going to hundreds of people that I didn’t GKH: I go around giving a lot of talks about you think it needs an upgrade, or do you re- know. But I really want to stress that it really who helps develop the kernel. At those gret choosing the GPL? isn’t about the innovation. The reason I’m talks, people ask how is helping here talking about Linux is, I think, the per- LT: I’m very happy with the GPL. I started out, and I say they’re very low on the list. A sistence and hard work of the kind of per- out with my own personal license that was lot of people found that interesting. They son who sticks to a project and does it every one paragraph, and it basically said “you didn’t realize that. My only response was day, 10 hours a day – there’s thousands of can charge no money for this, you have to that they weren’t contributing to the kernel. them now – and that was what’s brought give source code back,” and that was it. It They’re still not a very large contributor to Linux to where it is. It’s sweat and hard was probably not a license that would ever the kernel or most upstream source proj- work. stand up in court. The “no money can ects at all. Other people have verified that change hands” became a problem very Q: How will applications, in moving from as well. My only objection was, I want to early on. I looked around, and I thought the running on the OS on the desktop to run- see the kernel community grow. In order to ning on the web in a browser affect the de- grow, we need to have more developers velopment of Linux? contributing. Canonical has some very, very I’m very happy good developers. They are contributing LT: This has already helped Linux enor- more now. Over the past couple of years with the GPL. mously. I don’t know how it was in Japan, they’re doing much better – a lot more but both in and then in the US, you patches. had specialized applications, like for banking, GPL version 2 was exactly what I was look- that you had to run on Windows to talk to Q: What is the toughest technical problem ing for. I think it’s a very fair license, a li- cense that’s very successful, and I think it’s your bank. That’s all gone, obviously. Almost that you have faced during the develop- something that speaks to people at a very everything is a web application. That helps ment of Linux so far? deep level – the whole fairness notion that I Linux because, all of the sudden, the differ- LT: I’ll give two answers to that. They have give you something, you give me some- ences between operating systems aren’t as never been technical. Even when we make thing back. It’s worked very well. I person- important. When the differences aren’t as a wrong decision and take a wrong turn ally don’t care for the GPL version 3. I think glaring, now the technology matters a lot somewhere, eventually we figure out that it extended it to, “I give you something, more. Now the licensing and the price and was a bad decision and we can fix it. The you give me the code back, and you prom- just being available matters a lot more, and two areas where we’ve had serious prob- ise not to do certain things with it.” That that really helps Linux in the long run. lems was documentation and help from was never what I wanted to do. I’m very That said, I don’t think the web applications hardware manufacturers. Some hardware happy with the license. It’s clearly worked will take over the world entirely. I think it’s manufacturers have not been supportive, very well. Why change it? nnn

18 October 2011 Issue 131 linux-magazine.com | Linuxpromagazine.com

016-018_Linus.indd 18 8/10/11 3:39:33 PM