Ensuring That 2.6 Will Perform Better Than 2.4

Jeremy Andrews writes "Con Kolivas, a practicing doctor in Australia, has written a benchmarking tool called ConTest which has proven to be tremendously useful to kernel developers, having been designed to compare the performance of different versions of the Linux kernel. In this interview on KernelTrap he explains the project, noting that "a good 2.5 kernel (and that's not all of them) feels faster than 2.4 in most ways and this bodes well for 2.6." Also discussed is his high performance -ck patchsets, adding performance to the 2.4 stable kernel with the O(1) scheduler, kernel preemption, low latency, compressed caching and more..."

Comments

Beta testing 2.5!

[Blkdeath]

I certainly hope 2.6 out-performs 2.4, I mean, who'd want to take a step backwards, right? But we have a problem - the kernel team are implementing lots of top-notch functionality, but don't have enough people testing it. There are still compilation failures, ACPI buglets, and small quirks with several other sub-systems (speaking from my own experience here). Linus pleaded to the LKML not so long ago that he hopes lots more people will start plugging away so they can find and fix all the kinks before we go to 2.6 (or 3.0, like, whatever. :) ).

So download away and start testing!

Re:Beta testing 2.5!

[Blkdeath]

Hell, without improved USB and journalling filesystems, what was the point in 2.4?

The new VM was one of the big kickers (then they replaced it mid-stream with a better one, which was risky at best, but I digress). I've found that on all workstations and servers on which I've installed the 2.4 kernels performance is through the roof. It responds faster under all load conditions than any 2.2 kernel ever could.

I read the kt digest, but I'm really not seeing any new wonderful things. Perhaps there are others in a similar situation?

Pre-emption, new I/O scheduler, new paging, all sorts of wonderful stuff that makes the kernel more suitable for heavier loads. Most everyone who's tested it thus far sees it performing far better than 2.4 so far, and it's got a whole slew of improvements yet to come.

There's also talk of replacing LVM1 with LVM2, and supporting various other logical disk managers which will make it more desireable on the enterprise front.

compressed caching......

[oliverthered]

"compressed caching offers nothing to machines with heaps of ram",
If memory transfer and access speed is causing a bottleneck then well designed compressed caching can give a good performance increase by decompressing into the cpu cache.

Streems with small blocks would probably give the best performance increases.

Good stuff

[truth_revealed]

I'm very grateful that this guy took the time out of his busy schedule to quantify what every Linux user has suspected for a long time - CPU performance degradation during heavy IO. I have always felt that Linus put too much emphasis on pure CPU-bound tasks and that's why he resisted raising HZ above the ridicously low value of 100 for so long - to the detriment of desktop applications. Hopefully this a start of a trend to create a more universal general purpose kernel for interactive desktops, web servers and number crunchers.

Re:a good 2.5 kernel?

[cartographer]

The 2.5.x set of kernels is the development branch (the key being the odd minor number). It's where major feature changes are taking place, and new ideas are tried out. As you'd expect, things don't always work at the first (or fifth) go, so not every 2.5.x kernel is going to be 'good'.

'Good' probably depends a whole lot on how well the bits you need for your exact machine configuration are currently working, so your good kernel may be someone elses nightmare.

The lesson to all this is 'don't use a development kernel if your not ready for breakage'. That means back up all your file systems and don't even think of putting one on a production machine. But if you have some new widget that isn't supported in the stable kernel series (2.4.x currently), you might want to see how the development series is shaping up and even offer some feedback so that the next stable series really is.

Multitalented

[abdulla]

"Con Kolivas, a practicing doctor in Australia, has written a benchmarking tool called ConTest which has proven to be tremendously useful to kernel developers..."

Let me guess, he cooks too. Ladies this is your man.