by Tom Schaefer
netd@fastrans.net

TS: So what have you been up to lately?  

Linus: I've been working on various things: the current Linux SMP code and the new dcache are very heavily influenced by me. And apart from Linux I've had my RealJob(tm) to do, a family etc... 

In short, I certainly haven't been idle. 

TS: Are the rumors of your going commercial with the kernel true?  

Linus: Nope. Not at all. I actually haven't even heard such rumours, and I'm a bit surprised. The kernel copyright wouldn't really even allow me to make it commercial even if I wanted to, which I don't. And while I am the major copyright holder,  I have always encouraged people to put their own names on code they have written - as a result there are quite a few people who hold copyrights on different parts of the kernel. 

So everybody can sleep quite safely in the knowledge that it would take so many people to agree on making Linux commercial that it will never happen. 

And furthermore, even if Hell _did_ freeze over and some kernel would become commercial, that wouldn't take the GPL away from older kernels which would continue to work and be developed on by people who think it should be free. 

The end result is that there really isn't anything to worry about.

Linus: Right. Linux/SMP isn't very old, and the current stable kernels (ie the 2.0 series) have the traditional "first cut" of SMP - a single kernel lock. 

A single kernel lock to protect the integrity of kernel data is the simplest way to do a reasonable SMP, but it has the problem of not scaling very well on certain types of work-load (anything that is very system-call intensive, like a web server, for example). It works beautifully for many things (parallell makes etc), but it's not a really good answer in the end. 

The good news is that everybody knew that, and that the single-kernel approach was always meant to be just a first version. The current development kernels (the 2.1 series) already have a multi-lock SMP implementation, and it's coming along quite nicely. 

Currently much of the kernel is still protected by the single lock even in the development kernels, but the large difference is that it's no longer a conceptual necessity, but just the result of the slow migration from the original SMP model to the new one. 

So in 2.0 it was quite impossible to move any significant part of the kernel outside the lock, while 2.2 will already have many critical subsystems separated (process handling, interrupts, signals etc etc). 

TS: Many Linux projects underway  - do you see any particular project as most significant?  

Linus: If I had to pick one to work on myself, I'd still continue to just do the kernel, simply because that's where my personal interests lie. But I'm happy to say that these days most of the really important work is going on  in user space, and the kernel has been relegated to where it should be: a necessary part of the system, but people aren't thinking about it as much as they used to, simply because they are becoming used to having it there. 

Of the user space projects, I find wine (the MS Windows emulation stuff) and egcs (the follow-up on the original GNU C compiler project) to probably be the most interesting ones. But there are others that look  interesting, like the Corel NC that is Linux-based running on a StrongARM etc etc..

Linus: I personally use Red Hat, partly because I trust it technically. Red Hat has been very good about internet security issues, for example: as far as I can tell they are by far the most responsive vendor (and that's not just in the Linux world) when it comes to security issues. I'm essentially lazy, and it's just so easy to upgrade packages automatically over ftp when another sendmail security hole is found.. 

The other (and probably more pressing) issue for me is that I have three DEC alpha workstations at home, and Red Hat is one of the few Linux distributions that do multiple architectures. So that cuts down my choices quite a bit. 

However, I don't want to evangelize Red Hat - I do like SuSe a lot, for example. They have a really polished look, and while they were originally very German-based (their earliest CD's were all in German - I can decode the language if I have to, but I can't really read it), today they also do the normal English install. And partly because they've had the language problem in a way nobody else has had, they tend to be more aware of internationalization issues for example. 

And it depends a lot on what you need. Caldera is targeting a different audience than either Red Hat or SuSe, and because of their pretty desktop I installed that on my wife's machine originally (but then once she got used to it I switched her over to Red Hat only because I'm a lazy bastard and I had to have Red Hat on my alphas and I didn't want to administer two different systems if I could avoid it..). For some people the netware support in Caldera is going to mean that they are the only choice.. 

TS: You're in the US now? Can you tell us where you currently work, family, etc?  

Linus: Yes, I'm currently living in the US. Santa Clara, CA, to be more precise. We moved here early this year, and have been quite happy here. The weather is a lot better than in Helsinki, and I've enjoyed my work. 

("We" bein me, my wife Tove, our 10-month old daughter Patricia, and our two cats Randi and Misu. Yes, we took our cats with us when we moved). 

I work for a smallish startup called Transmeta, and the semi-official word is that we do "stuff". If anybody asks us about what kind of "stuff", the answer is "neat stuff". So now you know all the gory details. 

(But no, it's not actually Linux-related: I wanted to do something new at a place that allows me to do Linux as well as something else, and Transmeta is quite happy to have me continue doing Linux with no strings attached - possibly because they use it a lot internally ;)

(C) 1997 Tom Schaefer Net Design Technology, Inc.