2.6.8.1-nitro5 "it cleans your breath while it cleans y

[Pink]

If you don't mind me asking, what do you use in place of Win4Lin now? Or did you need for something like that go away.

Anyway, thanks for your help PickledOnion.

No I don't mind you asking - I needed win4lin for a neural network simluation programme I used at University. I've finshed uni now and have no need for any windows progs (well, except for any gaming I do and then I dual boot into win2kpro).

I won't be carrying on the patches simply due to lack of time: I was away for two days and found 9 kernels that would 'require' a win4lin patch if I was to carry on with my website (from the mm's, to rc's and rc-mm's, nitro, love, etc).

I simply cannot commit the time needed to provide a quality patch - As it is only one patch, I ensure it is fully working on a number of systems and with different distro's. I enjoyed it but have no real incentive to carry on and, as I say, I really cannot commit the time needed.

Thanks for the kind words though

Oh, and in case you were wondering, I have gone back to 2.6.7-ck5 for my kernel. It is simply the quickest, most responsive and most stable kernel I have used, with nitro5 a close 2nd.

[tatesworld]

tried this kernel,
very responsive

only trouble, its no good for cedega, Soldier of fortune 2 had cutting out sound with this kernel when my brother tried.

[teutzz]

only trouble, its no good for cedega, Soldier of fortune 2 had cutting out sound with this kernel when my brother tried.

you have to renice cedega/wine/winex to +19 (you've read right +19) maybe it sounds absourd, but just give it a try

[Robin79]

only trouble, its no good for cedega, Soldier of fortune 2 had cutting out sound with this kernel when my brother tried.

you have to renice cedega/wine/winex to +19 (you've read right +19) maybe it sounds absourd, but just give it a try

And if you use p2p (Point2Play)?

[ryceck]

I have to admit I'm kind of a love-sources junkie but this one made me change my mind
The kernel is stable and works good for me

Just need 1 more thing:
-The fbslpash that is here... how do I get bootsplash working with this..?

Just keep up the good work

[jewps]

fbsplash is a bootsplash replacement, there is a nice guide over at http://dev.gentoo.org/~spock/

its under gensplash.


sooo who's going to make the 2.6.9-nitro?

[Rainmaker]

only trouble, its no good for cedega, Soldier of fortune 2 had cutting out sound with this kernel when my brother tried.

you have to renice cedega/wine/winex to +19 (you've read right +19) maybe it sounds absourd, but just give it a try

It's true... This helps... I figured that out just before I read this post... I tried -20 too, but sound get's even more choppy then. I also deselected OSS and am running pure ALSA now.. That seems to help a bit too.

I'm runnning this kernel now too. I got a slight "problem". My harddisk activity is much higher then it just to be. I'm running azureus for example. Memory and CPU usage is very low, but my harddisk keeps writing, even though I'm only downloading at +- 100 kb/s. It starts using a memory buffer, but then only writes a few kbs. Is there any way to control the amount CFQ writes to disk in a pass? Or is there another scheduler which fills up my buffers, then writes out about 50 Megs at once?

[miseiler]

I'm runnning this kernel now too. I got a slight "problem". My harddisk activity is much higher then it just to be. I'm running azureus for example. Memory and CPU usage is very low, but my harddisk keeps writing, even though I'm only downloading at +- 100 kb/s. It starts using a memory buffer, but then only writes a few kbs. Is there any way to control the amount CFQ writes to disk in a pass? Or is there another scheduler which fills up my buffers, then writes out about 50 Megs at once?

Kernels based on the -ck4 patchset (such as nitro5) have a bug described by Con Kolivas as follows:

Code: Select all

I know I"m meant to be taking a break but I modelled in my head a 
problem with the mapped_watermark patch. It is currently too aggressive 
and the flags can get mixed up leading to an easy out-of-memory 
condition, and too aggressive cache cleanup which may slow down 
application startup.

Basically, the system is using your swap (disk thrashing), even when there is plenty of free RAM left.

He goes on to mention two patches that he believes would help the situation, but three of us following this thread have tested them and, if anything, it makes things worse.

If you're having this problem, I suggest moving back to nitro3. It was never officially released, but of all the nitro-sources it's been the very best for me in terms of stability (especially with reiser4) and performance.

[Rainmaker]

well, I don't have out of mermory conditions, although my "swapiness" seems to be a bit high indeed.

Code: Select all

dannie@Medusa / $ free -m
             total       used       free     shared    buffers     cached
Mem:          1009        995         14          0          0        903
-/+ buffers/cache:         91        918
Swap:          287         54        233

It's swapping 54 megs, even though there's 918 Megs of RAM left...

I can even run max payne 2, xmms, firefox and the gimp al at the same time with no noticible lag... It's just annoying to see that red light blinking

I'll just wait for the next patch. Couldn't this be resolved by switching to the no-op scheduler or another i/o-scheduler?

[Bot24]

2.6.8.1-nitro5 is extremely slow, messes up my first USB mouse(I have to unplug both mice and then plug them back in), and gets my display stuck at 800x600. Athlon XP 2400+. 1GB RAM. Radeon 9800Pro(drivers installed)

[Rainmaker]

have you enabled "support for hot plugging devices"? Does hotplug start when your system boots?

And what do you mean "sticks to 800x600"? Also in X? Or just in console? You can configure the default resolution, but I think you need vesa lines in grub/lilo.conf to change that

[bisho]

The people who is having problems with swapiness should try to find the proper values changing vm.swappiness and vm.vfs_cache_pressure in the /etc/sysctl.conf.

vm.swappiness goes from 0 (none) to 100 (always)
By default is arround 60, but I have found to be better at 30 or even less for desktop usage, so you avoid all the trashing when using again an app not used for some time. In servers 60 is a good value.

vm.vfs_cache_pressure goes from 0 (try to make cache always) to large numbers (try not to cache). By default is 100. Depending on your amount of ram increase or decrease.

In my laptop, with slow disk (slow swap) I use swapiness of 20 when using desktop, and 60 when compiling and not in front of computer, and vm.vfs_cache_pressure to 200.

You could also change the values on the fly, by using sysctl:

Code: Select all

$ sysctl -w vm.swappiness=30

or the /proc interface:

Code: Select all

$ echo 30 > /proc/sys/vm/swappiness

[Bot24]

I don't know about the hotplugging, but it works fine with the Gentoo sources. My second mouse and my Bluetooth adapter are USB also, but they don't need to be disconnected and reconnected(I was reseting the second mouse because the order was wrong).
800x600 in the console and X it looks like. Bootsplash didn't start, but that is probably my fault, because it doesn't work on the last Gentoo kernel I built either.

[Rainmaker]

Thanks for your insight, but I get this:

Code: Select all

root@Medusa root # sysctl -w vm.swappiness=20
error: 'vm.swappiness' is an unknown key

Code: Select all

root@Medusa vm # pwd -P
/proc/sys/vm
root@Medusa vm # ls
block_dump                 laptop_mode            nr_pdflush_threads
dirty_background_ratio     lower_zone_protection  overcommit_memory
dirty_expire_centisecs     mapped                 overcommit_ratio
dirty_ratio                max_map_count          page-cluster
dirty_writeback_centisecs  min_free_kbytes        vfs_cache_pressure
root@Medusa vm # 

Might try playing around with the dirty values though. That was actually what I was looking for in my first post Thanks.

[y0zza]

Well, I reversed the mapped_watermark patch, so the kernel is back to the 'plain' cache/swap algorithms, and things have improved a lot - back to normal now:

Code: Select all

blah vm # free -m
             total       used       free     shared    buffers     cached
Mem:          1518       1481         37          0         16        842
-/+ buffers/cache:        622        896
Swap:         1265          0       1265

(this is without mapped_watermark at all)
Compare to with mapped_watermark:

Code: Select all

blah y0zza # free -m
             total       used       free     shared    buffers     cached
Mem:          1518        714        804          0          1        385
-/+ buffers/cache:        327       1190
Swap:         1265        113       1152

It does seem like an issue with highmem and mapped_watermark, at least in my case (1.5GB RAM), as even with the revised patch, the kernel seemed very reluctant to cache file data, even with the mapped 'watermark' set to 10 (default 66), almost as if it was only caching in highmem. 600+MB of RAM was always available (all from lowmem), no matter what I tried to do; the system was even happy to page 200MB+ of things to disk, with 700MB physical RAM free.

Anyhow, with the mapped_watermark.diff reversed, nitro5 seems quite responsive on my system, since there's no longer any of that quirky cache/swap behavior.

[Akuma]

Thanks for your insight, but I get this:

Code: Select all

root@Medusa root # sysctl -w vm.swappiness=20
error: 'vm.swappiness' is an unknown key

Code: Select all

root@Medusa vm # pwd -P
/proc/sys/vm
root@Medusa vm # ls
block_dump                 laptop_mode            nr_pdflush_threads
dirty_background_ratio     lower_zone_protection  overcommit_memory
dirty_expire_centisecs     mapped                 overcommit_ratio
dirty_ratio                max_map_count          page-cluster
dirty_writeback_centisecs  min_free_kbytes        vfs_cache_pressure
root@Medusa vm # 

Might try playing around with the dirty values though. That was actually what I was looking for in my first post Thanks.

have the same problem when trying to change swappiness.

[zerojay]

I've tried a lot and I still completely hard lock whenever I run emerge. I had this same problem with earlier versions of nitro. Seems like the kernel dies on "enqueue_task", I believe. Oh well.

[discomfitor]

I think I fixed the problem (slowness)...not sure though, so attempt the use of this patch at your own risk.

[miseiler]

Well, I reversed the mapped_watermark patch, so the kernel is back to the 'plain' cache/swap algorithms, and things have improved a lot - back to normal now:

Code: Select all

blah vm # free -m
             total       used       free     shared    buffers     cached
Mem:          1518       1481         37          0         16        842
-/+ buffers/cache:        622        896
Swap:         1265          0       1265

(this is without mapped_watermark at all)
Compare to with mapped_watermark:

Code: Select all

blah y0zza # free -m
             total       used       free     shared    buffers     cached
Mem:          1518        714        804          0          1        385
-/+ buffers/cache:        327       1190
Swap:         1265        113       1152

It does seem like an issue with highmem and mapped_watermark, at least in my case (1.5GB RAM), as even with the revised patch, the kernel seemed very reluctant to cache file data, even with the mapped 'watermark' set to 10 (default 66), almost as if it was only caching in highmem. 600+MB of RAM was always available (all from lowmem), no matter what I tried to do; the system was even happy to page 200MB+ of things to disk, with 700MB physical RAM free.

Anyhow, with the mapped_watermark.diff reversed, nitro5 seems quite responsive on my system, since there's no longer any of that quirky cache/swap behavior.

Excellent work. I have taken your idea just a small step further in my own testing, adding the hard_swappiness1.diff patch from 2.6.7-ck6 at the suggestion of Javier Marcet on the ck mailing list.

The results (so far) are fantastic. I very highly recommend to anyone using the nitro5 kernel to reverse the mapped_watermark.diff patch and add the hard_swappiness1 patch in its place. I've spent ten minutes copying, taring, and untaring large movie files while running this kernel and thus far the swap space hasn't been touched.

Do yourself a favor and give this a shot.

[Bot24]

I'll try again with the plain VESA(not the tng) and the reversed/added patches. I still don't know about that mouse strangeness. My tablet(mouse #2) doesn't work under X right now. It has all of the drivers loaded, and the Xorg logs confirm it(xxd and wacdump too), but nothing happens when I move anything on it. I think I just remembered something tho, and I will go digging in the source code to look for alternatives to stylus... I think that might have been it.
I would prefer to use a real mouse, however, so if anyone knows why one USB mouse will escape detection on boot(but not on insertation) and the other won't, please tell me.

[neonik]

Have you given the new watermark v2 and updated staircase a try?
It's in the root of 2.6.8.1: http://ck.kolivas.org/patches/2.6/2.6.8.1/

And that swappiness issue, I guess it's handled via mapped. You can nevertheless just use the echo command: echo 20 > /proc/sys/vm/swappiness

Furthermore there's a new patch in -ckdev, which I got no idea what is for: http://ck.kolivas.org/patches/2.6/2.6.8.1/2.6.8.1-ckdev (sc_mw.diff).

Edit: I've just taken a look at it, what I believe is that it's meant to solve the mappins/wap issue.

Here direct links:
from_2.6.8.1_to_staircase8.1
mapped_watermark2.diff
sc_mw.diff

[Bot24]

This is starting to sound complicated. Mostly because I don't know who people are addressing, I think. What should I do? Just reverse the watermark patch and then apply the hard swappiness patch?

Wow. "swappiness" is in the spellchecker. Why not USB and such?

[teutzz]

Have you given the new watermark v2 and updated staircase a try?
It's in the root of 2.6.8.1: http://ck.kolivas.org/patches/2.6/2.6.8.1/

And that swappiness issue, I guess it's handled via mapped. You can nevertheless just use the echo command: echo 20 > /proc/sys/vm/swappiness

Furthermore there's a new patch in -ckdev, which I got no idea what is for: http://ck.kolivas.org/patches/2.6/2.6.8.1/2.6.8.1-ckdev (sc_mw.diff).

Edit: I've just taken a look at it, what I believe is that it's meant to solve the mappins/wap issue.

Here direct links:
from_2.6.8.1_to_staircase8.1
mapped_watermark2.diff
sc_mw.diff

i've patched the kernel (nitro5) with staircase 8test1, watermark fix, and watermarmark2 and i must say that it feel much better, even as for swaping is concearned
i tried to patch it also with 8.1 staircase but it gives me some failed chuncks and i have no ideea (in genertal speaking) how to fix hailed chunks so i left this patch out
maybe later today i'll give a try to sc_wm

[Vide]

Strange, my radeon card worked fine. It might sound silly, but you HAVE emerged ati-drivers then opengl-update ati since you installed the new kernel, right? And are you using the 3.9.0 drivers or the 3.11.1 drivers, which work for me?

What does fglrxinfo show?

Obviously I reemerged and updated everything, I'm not in mesa
The nly thing is that I'm using 3.9.0 ati driver, and I'm running a NForce2 mobo. What's your chipset?

Anyway I'll give a try to older ati driver version

[neonik]

teutzz: you gotta reverse staircase 8.0 before applying the new one.

As for sc_mw "(scan control for mapped watermark)", look here: http://bhhdoa.org.au/pipermail/ck/2004- ... 00599.html and http://bhhdoa.org.au/pipermail/ck/2004- ... 00600.html