Con Kolivas' kernel patchset discontinued

[incognu]

Con Kolivas is discontinuing his fantastic kernel patchset as of 2.6.22-ck1: http://ck.wikia.com/wiki/End_of_ck_patchset

You can find the patches here: http://members.optusnet.com.au/ckolivas/kernel/

In addition to patches for 2.6.21 and earlier kernels, there are also patches for 2.6.22 rc kernels.

This is unfortunate ... I've been using his patches for a while, and find they enhance desktop performance.  If you want to try them, they're easy to apply.  I think he has a "walkthrough.txt" in the patch download area.

[easuter]

I wasn't even aware there was an alternative kernel scheduler! 
Well, its certainly sad to see that its maintainer left/abandoned the project, seems like it had lots of fans, and from the comments on the mailing list, it seems very promissing performance-wise!...

[incognu]

It really helps tweak performance. 

I usually can't tell I'm compiling something when I use schedtool with his patches.  The desktop feels as smooth and responsive as ever.

I don't know how well the CFS scheduler will work; it's apparently somewhat based on Kolivas' Staircase Deadline ...

I hope someone will maintain the patchset -- which includes other patches besides the scheduler.  It's possible, considering the many people who are so enthusiastic about it.

[easuter]

I hope someone will maintain the patchset -- which includes other patches besides the scheduler.  It's possible, considering the many people who are so enthusiastic about it.

I hope so too...
I'm gonna give it a try as soon as I get a chance (when my *real holidays* start, anyway )

[easuter]

Well today I installed SOHO on one of my "go wild, doesn't matter if you screw it up" partitions  and compiled 2.6.21 with the -ck patchset.
I chose SOHO, because since its the slowest on my box, its where I would feel a difference in performance, if there were to be one.

From the grub boot screen to the KDM login splash, the patched kernel shaved 6 seconds of boot time, compared to vanilla SOHO's boot time (yeah, I used a stopwatch ).

After that, KDE loaded almost right away, significantly faster than without the patches. Same goes for Seamonkey, which is normally like molasses on my box, and big fat tarballs with -9 compression...the terminal output spews out much faster than usual.

I think I'm gonna use this for my packaging partition....

[Triarius Fidelis]

I usually can't tell I'm compiling something when I use schedtool with his patches.  The desktop feels as smooth and responsive as ever.

How much better is it? I lag slightly at times when compiling something like ghc.

[exeterdad]

Thanks for the info easuter.  I think I'll patch mine as well.  My old box needs all the tweaks it can get.

[incognu]

I usually can't tell I'm compiling something when I use schedtool with his patches.  The desktop feels as smooth and responsive as ever.

How much better is it? I lag slightly at times when compiling something like ghc.

Subjectively, much better.  For example, when compiling something like the kernel or seamonkey, and browsing in seamonkey at the same time, I'd normally notice some lack of responsiveness.  When using the ck patchset and schedtool -D e with make, there's no subjective lag, no real indication that a build is going on.  (Doing a speedtest, like Scragz' page load test, shows a slower page load time, but it's not something I'm subjectively aware of.)

I've also aliased tar to use the IDLEPRIO scheduling, and bz2 is much pleasanter now!

According to CK's walkthrough.txt, using schedtool with a -ck kernel will cause the process to use only spare cpu cycles.

Easuter:  glad it's working for you!  VL is already fast and responsive, but the ck kernel does provide an extra tweak.  The improvements aren't limited to providing a more pleasant compile experience, by any means! 

[easuter]

Well, I'm going to package this and schedtool for the repo.

Since 2.6.21 has a tickless timer option, how does everyone feel about having that enabled too? 

[incognu]

Well, I'm going to package this and schedtool for the repo.

cool!

Since 2.6.21 has a tickless timer option, how does everyone feel about having that enabled too? 

I think it's a good idea!  I've been using it, and thought it might be a good idea to enable it by default in VL kernels.  CONFIG_HIGH_RES_TIMERS as well as CONFIG_NO_HZ?

[easuter]

Are you using both, or just CONFIG_NO_HZ?
I haven't tried either (or both) yet, and probably won't be able to for a couple of days...

[incognu]

Both.

Hmm, I was just re-reading http://kerneltrap.org/node/6750 and saw that CK has/had a "dynticks" patch with similar functionality.  I'm going to grep a bit ...

[edit] Nope, ignore the dynticks comment.

I've read some really positive comments about the tickless kernel.  I've been using it for a while: nothing spectacular that I've noticed, but it seems worth turning on. 

[easuter]

Code:

I've been using it for a while: nothing spectacular that I've noticed, but it seems worth turning on.

Well, I was mostly interested in the tickless timer option because it should reduce power consumption.
Can't remember where I read it though, that this addon for x86 is part of an attempt to make the kernel "greener" and also perform better on laptops.

Apparently its been available for the S390 for a number of years already, and the x86 port has been available for about a year, but wasn't mainstreamed until now.

Oh well, I'll enable both options in the -ck patched kernel, if nobody has any objections to that

[incognu]

I think I read something about a "greener" kernel too, but can't recall; it was in the past couple of weeks.

You've probably seen it, but GrapefruiTgirl started a discussion about the tickless kernel over at LQ.  There were some comments that it could improve responsiveness, as well as reducing temperatures and energy consumption.

[easuter]

Unfortunately, the tickless timer and high resolution timer options make my computer lock up, so I guess old hardware does not support these features :\