Linux 2.6.28 Kernel Benchmarks

[Michael]

I really like what you are doing here, by carefully benchmarking every kernel and whatsoever, but I think the data could be presented better.
Right now there is 2-3 pictures per page with a few paragraphs of text spanning over 6 pages, not mentioning that the graphs don't show anything that really needs graphs. It's frustrating to keep going over every page just to see something that could be summed up with 1 or 2 sentences.
Perhaps you could make the graphs more compact in the future, because now they take up too much space and are bulky. Also, I think it is safe to allow more information per page, because now there's atleast half of the page height still empty white, unused.
Thanks

We have costs that need to be covered somehow... Hence advertisements and Phoronix Premium. Premium subscribers can click a single button and view all six pages on a single page.

[Luis]

It really proves how useful it is to have a performance testing infrastructure in place (the kernel is thoroughly benchmarked on every RC, by Intel and IBM people especially) and how it prevents regressions. If only Xorg could do the same...

[bulletxt]

@ hoho

If you can't click 6 times to see the full benchmark, then just don't turn on your PC and surf the net. Also considering this website has interesting Linux articles and it's free, Michael does a really good job with advertisment not being frustrating and I appreciate this...and we should all appreciate this.

About the benchmark, I'm glad to see there aren't regressions and this isn't something obvious. I know this can sound stupid, but I'de be curious to do the same benchmark running different DM, for example running KDE4, XFCE and stuff like that just to see if there are some serious differences running one or the other. I hope Michael can someday do this as I would really appreciate it.

[gamesfan]

Why not using ext4 to mount the ext3 partition?

Although I do not think the whole test is worthless, it would have been MUCH more interesting if you were mounting the ext3 partition with ext4 module - which is possible thanks to ext4 backward-compatibility and still is supposed to provide some of the performance improvements of ext4.

Any chance you update your test with this?

[curaga]

Weren't those performance improvements only for newly created files?

[TOGGI3]

umm, do the nvidia drivers even make use of GEM at all?

if not this test is seriously lacking relevance to the inclusion of GEM in the kernel...

Someone needs to do an old kernel and new kernel test on intel graphics I think.

[Michael]

umm, do the nvidia drivers even make use of GEM at all?

if not this test is seriously lacking relevance to the inclusion of GEM in the kernel...

Someone needs to do an old kernel and new kernel test on intel graphics I think.

NVIDIA's drivers don't use GEM, but this article isn't a test of GEM at all but the overall kernel performance.

[pking55]

Registered just to post this:

FPS in wine dropped sharply with 2.6.28, making it almost unplayable. I reverted back to 2.6.27.10 for now. All I play is hl1 based mods, so you might want to try benchmarking those.

Also I noticed that sound messed up every one in a while while watching videos. A quick pause/play fixed it though.

Could you post a link to this thread? I can't seem to find it.

[pking55]

@hybrid-kernel

It's wine problem. Maybe you'll have to tune Linux scheduller especially for wine to have better experience in games. It's mentioned at wine forum.

Could you post a link to this thread? I can't seem to find it.

[Kano]

I don't think many use the latest kernel because it is faster. It is just newer, "feels" better and for those who had to compile extra drivers those could be intergrated already (for default distro kernel users this last aspect does not really matter, because drivers are always added when needed). Sometimes newer kernels are required, especially for newer hd controllers, which might not be supported before. Of course some might want to use the extra features like KMS or whatever, but you will never see that in a benchmark.