You were recently at the Linux Kernel Summit. Can you tell us a little bit about the summit, what it is, and what you saw?

The Linux Kernel Summit takes place every year and the last few years it's been held in Ottawa, Canada. Prior to that it was in San Jose, California. USENIX organizes the whole event and they're sponsored by various companies; Oracle is a big sponsor,as well as companies like IBM, HP, and Intel. It's mostly an invitation-only event. There's a small committee with five or six members that decides who should go based on influence or feedback that people can give toward addressing the kernel team. In general, there's 70-80 people, pretty much the core Linux Kernel team. Most of the developers who you would see on the Linux Kernel mailing list, most of the people who do contributions to the kernel are there. Most of the vendors are represented in some way. There's lots of folks from Red Hat, IBM, Oracle, HP, Intel, SUSE, and so on. Basically all the big players are represented.

It's similar to an architecture meeting with everybody present, which is better than only discussing issues by the mailing list. In addition, since the folks are right there, people go off and talk to each other in the hallways and have separate meetings to discuss what they're working on. So it's very productive. It's only two days, but it's two very long days. It starts at about 8 or 9 a.m. and lasts until 9 or 10 p.m. That is the overall reason behind the Kernel Summit.

It sounds like a really interesting summit. I was wondering, since you have representatives from all these different companies, do you find that the people who go there are pushing the agenda of their individual company? Or how does that work?

A little, of course. If you work for a company like Oracle, you know what your customers are facing and you know what sort of features they need. Part of the feedback you can provide will be based on your experience. It's not just a way for the corporate world to give their features to the Linux community. That's not what it's about. These people have been using Linux in different industries for a long time; when they present their findings, because of their line of work, it lines up with the companies they work for. I think that's a better way of expressing it. I don't think it's fair to say that whoever works with company XYZ comes there with a list of proposals that that company asks for. There is some of that, but I don't think that's a real challenge for them.

In addition to everybody coming as a member of a company, they also focus on what's important for the larger Linux community.

Exactly.

What were some of the big new features, or what were some of the major topics of discussion?

This year was a little different because, for one, everybody is very, very happy with the abilities of Linux 2.6. And everyone is very happy with the development model, which is a little bit new this time, and the number of features that are there. In short, Linux 2.6 contains most of the stuff that everyone needs. It's still lacking some features, but in general, everyone is pretty happy with it.

Based on the discussions we had, there is no real need to open up a 2.7 tree. The idea is to just continue with the 2.6 model, add some functionality and not worry too much about 2.7.

One issue is that the vendor kernels deviate too much from the standard kernel. If you look at what Red Hat ships or what SUSE ships, there are tons of patches that they apply on top of the standard Linux kernel. And, unfortunately, not enough of those changes go into the main standard kernel fast enough, so you end up with vast differences between Red Hat's tree, the Linux tree, and the SUSE tree.

That's a real pain in the neck for application vendors like Oracle and IBM. If we build Oracle and we build it against one or another of the distributions, there are so many changes between them that we have to do tons of testing. It would be much better if we actually had one single tree that all the distributions were really close to. There's a push now to be as close to the main line as possible.

Given the reasonable stability of the 2.6 kernel, what do you think that means for Linux users and companies adopting Linux? Does that mean they'll be with 2.6 for a relatively long time compared to some of the recent kernels?

One would hope. If I were to write down a list of whatever you have in Linux 2.6, and whatever customers need, I think that pretty much all of them are there. One would hope that with 2.6, the release cycles will slow down and it will have more value going forward and have a longer release cycle. That's what it sounds like.

About Wim Coekaerts
Wim Coekaerts is director of Linux engineering at Oracle.
wim.coekaerts@oracle.com