Related link: http://kerneltrap.org/node/view/3513
Recently the news came out (mostly from this kerneltrap article) that the 2.6 kernel tree was to continue to receive active development. Torvalds and Morton (the current 2.6 maintainer) have noted good results in incorporating patches into the 2.6 tree that in the past would have gone into a development tree. Because of this, they have decided to keep with the 2.6 tree for the foreseeable future, and only announce a “development” 2.7 tree when enough patches are submitted that could seriously disrupt the kernel’s stability.
Linus, who normally can do no wrong, received quite a bit of flack back in the middle of the 2.4 series when he allowed a completely different virtual memory manager to be dropped in. Linus argued that the code was good and the improvements were needed, but sysadmin who had finally trusted 2.4 enough to migrate servers to it from 2.2 complained that any major development should have gone in 2.5. That combined with some fairly serious stability problems in certain 2.4 releases (2.4.11 comes to mind) have made quite a few administrators, myself included, a bit gun shy about the “stable” kernel series.
As a sysadmin, one of my main goals is to ensure that the servers I run stay up. Unless I’m forced to upgrade due to an exploit, some of my servers might be a few kernel revisions behind, as it simply isn’t worth downtime just to have the latest kernels. Currently, all of my machines are on the 2.4 series, along with some of my desktops. I have started to play with 2.6 on a desktop or two since around 2.6.3 or so, but I still am a bit reluctant to use it on many machines, and definitely not any production servers. It’s a shame, because the 2.6 series is definitely faster, works better with ACPI, and has a lot of other interesting features.
In many ways, I have always looked at the early releases of the “stable” kernel series a bit like I look at Debian testing/unstable. Most of the time, Debian unstable packages are pretty solid (I’d argue often more stable than the packages other distributions put on their *.0 releases), and even when they aren’t, if you have some Debian experience, you can generally fix the problems that do crop up.
I love Debian unstable, and I use it on all of my desktops, however I don’t use it on any of my production servers. I’m starting to get a similar feeling about the 2.6 kernel series. Before this shift in the way Linus and Morton handle patches to the stable series, I was planning to start the process of migrating servers to 2.6 “after it stabilized” ie. some time after 2.6.10 or so. Now, I’m really not so sure.
The way it appears now, the 2.6 series will be treated like Debian unstable–use it, but you should know what you are doing, and if you use it on a server, you deserve what you get–or possibly a blend between that and testing. The 2.7 kernel, or the -mm series for 2.6 will be more like Debian experimental.
So, when will you deploy 2.6 on your servers? On your desktops?