2.6.11-no1: Underground polar bears in the artic!

[joecool]

This is a bit of a basic release, but its out there to be stable. The next release is slated to have alot more features as I have quite a few people coming together to do good things. Our intent is to make mm-sources runnable, while adding other features that may help improve performance, or add support for drivers.

This release was mostly done by predatorfreak, and was completed by myself. Next release will probably feature additions from blutkind and bzcat_-|patch_p1. We are still working on a website which is now in development.

I realize I may recieve a few complaints that feature X is missing from this release, we will be sure to include more in the next release. We will NOT include genetic schedualing for a few reasons. The main reason being, although I see it as interesting technology, my tests with it have proved far from impressive. In the future if it can adapt fast enough to system changes, I will add it. I don't see that in the near future however and will not have my performance or time wasted putting it in. Vesafb-tng is something many users are a little crazy over at the moment. I see it as a waste of time, but depending on if it decides to work or not, it might be in the next release.

The current release can be fetched at http://no.oldos.org/files/2.6.11-no1/. Its based on 2.6.11-mm4 and there is an ebuild. Patchlist goes as follows:

[bzcat_-|patch-p1]

Yeah nice to see a stable release in the mix as opposed to our crazy wild releases.

[voidengineer]

Very nice!

Thanks joecool

voidengineer

[bzcat_-|patch-p1]

Hmm certainly an image that will stay in my / boot/ partition. =)

[joecool]

I'd also like to announce my results with different IO schedualers on reiser4.

I've long been a cfq zealot.. but those days are behind me now. I wanted to do a real life test, that unlike fallow's would actually test something that has annoyed me for quite awile.

How this benchmark was done.. is not exact science, I tried to put my box through one of the more grueling times of IO usage. So, in order to do this, i simulaneously deleted a kernel source tree while unpacking another. I played music at the same time through alsamixer at around 192kbps.

The results of this test were not times.. i wanted to measure interactivity as best as I could. The results were suprising.

Cfq, which is supposed to be fair queing, was HORRIBLE. Music skipped left and right, and there were points when the mouse would barely move across the screen. Not to mention it took a mad long time to finish unpacking and deleting. I can't say I recommend it for ANY work that involves lots of IO activity. I've not tested cfq-ts.. but i will in my next release.

Deadline, I was quite suprised on. It finished the test much faster then CFQ, by far it has the fastest throughput. On the downside, while Deadline has better interactivity then CFQ, music still skipped occasionally, and you could definetly feel things bog a little. Overall, I recommend it for anyone who wants to run a server, anyone gaming, and anyone who needs to move files very fast. For fileservers, nothing can match it.

Anticipatory, I've never been a huge fan of this.. but this test has come to make me like it alot more. Anticipatory was not very fast, I didn't expect it to be, but it did seem on-par with CFQ's speed. On the other hand, anticipatory GREATLY accels on one thing. Interactivity was amazingly good, even while stressin the system with this test and an emerge sync on top of it, the music skipped twice, and it was barley noticable. A rock solid IO sched I recommend for the average Desktop user. It performs well for everyday use, but I recommend goin over to Deadline if you don't need the interactivity as much, as the performance will fill the gap.

I did not test no-op, but from what other maintainers have been telling me, its meant for RAID configurations, of which I do not have, so I won't test.

On a final note, the genetic anticipatory had high throughput, but would let processes that didn't need as much IO starve.. and it would take 5-10 seconds for one to pick up. Once again.. no suprise.

Fortunatly IO scheds are easilly picked in /sys/block/hdx/queue/schedualer.. just echo deadline, anticipatory, cfq, or noop.

[predatorfreak]

[joecool]

as i said before.. i did not have the chance to test cfq-ts.. but the results from cfq were not too impressive. I'll be sure to test it with my (or our) next release.

[predatorfreak]

[bzcat_-|patch-p1]

As I have stated I'd go with deadline if not anticipatory as default. But I do run reiser4. Perhaps some test with other filesystems wouldn't be a bad idea.

[joecool]

suppose i could do tests on XFS.. i do have it on my backup install.