The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

pankajmore · 2010-09-28 19:14:26

this kernel is awesome , the system feels very snappy , but sadly hibernation doesn't work , after hibernation , during "..kernel and process data ..." the caps lock led continuously blinks and i have to manually power down by pressing the power button , kernel26-ice works perfectly , so tuxonice works fine on my hardware , does anybody else has issues with hibernation??

nous · 2010-09-29 16:15:59

pankajmore wrote:

the nvidia-pf package from dropbox repo , throws invalid module error with the pf kernel from the repo, anyone else having such problems , please confirm...

Which repo, i686 or x86_64? If the latter, the prebuilt nvidia-pf is intented for the generic x86_64 kernel, not the optimized ones.

pankajmore wrote:

this kernel is awesome , the system feels very snappy , but sadly hibernation doesn't work , after hibernation , during "..kernel and process data ..." the caps lock led continuously blinks and i have to manually power down by pressing the power button , kernel26-ice works perfectly , so tuxonice works fine on my hardware , does anybody else has issues with hibernation??

Again, which repo? If 64bit, is it the generic x86_64 or the core2 one? I think I had similar trouble with core2 on a core2 duo server at work, but in my core2 laptop works just fine. But then again the laptop package is custom-built for it, not the generic core2 one and I don't usually hibernate the server, so I can't really test.

Therefore, I would blindly advise you tweak the /etc/hibernate/common.conf...

[EDIT]
And I really don't know why I ever assumed you have a Core 2 CPU...

[EDIT2]
The -pf9 patchset updated the BFS patch to version 250, maybe that'll help. You have to compile the kernel26-pf from the AUR though, It'll take me some time to update the repos.

Last edited by nous (2010-09-29 16:35:51)

pankajmore · 2010-09-29 16:37:33

@nous its a core i7 laptop ,nvidia-pf problem is solved, and i use the x86_64 core2 optimised kernel , i use pm-utils , should i switch to hibernate-script although it has the same problem...One thing that is peculiar about its hibernation is after "...kernel and process data.." the caps lock led blinks , but when i power down and reboot , it resumes successfully , which implies that hibernate works fine , only that it doesnt shutdown/halt after hibernation and throws led blinks at me , Besides, , the repo has pf8 whereas aur has pf10 , should i continue using the repo , or change to AUR , it ll be nice if you do keep updating the repo though...

Do you also put the Xorg ISO_Scheduling patch , it seems to improve the responsiveness of xserver? If its not there in the default PKGBUID , i ll try to manually patch it....

Last edited by pankajmore (2010-09-29 16:44:49)

Cdh · 2010-09-29 17:32:15

Hm... I use the pf10 i686 kernel at the moment, did build it from aur.

I use an Athlon II 240 and use make -j2.
While compiling another kernel every few seconds I get a big input lag for ~1 second and everything else stalls...
All the time htop shows that gcc uses nearly no cpu, but the CPUs seem to be much used.

Last edited by Cdh (2010-09-29 17:32:45)

nous · 2010-09-29 17:37:27

pankajmore wrote:

@nous its a core i7 laptop ,nvidia-pf problem is solved, and i use the x86_64 core2 optimised kernel , i use pm-utils , should i switch to hibernate-script although it has the same problem...One thing that is peculiar about its hibernation is after "...kernel and process data.." the caps lock led blinks , but when i power down and reboot , it resumes successfully , which implies that hibernate works fine , only that it doesnt shutdown/halt after hibernation and throws led blinks at me , Besides, , the repo has pf8 whereas aur has pf10 , should i continue using the repo , or change to AUR , it ll be nice if you do keep updating the repo though...
Do you also put the Xorg ISO_Scheduling patch , it seems to improve the responsiveness of xserver? If its not there in the default PKGBUID , i ll try to manually patch it....

I try to upload the latest packages as fast as I can, but real life has to take precedence. I'm compiling the i686-pf10 right now, the 64-bit versions should be up by tomorrow evening.

Since time is a constraint, I try to keep the included patches few and maintainable. The -pf patchset itself anyway contains almost everything a real-time fan could ask. And, to be honest, if an i7 isn't snappy enough on -pf alone, then no tweak can help you ;-)

pankajmore · 2010-09-29 18:19:40

nous wrote:

pankajmore wrote:
@nous its a core i7 laptop ,nvidia-pf problem is solved, and i use the x86_64 core2 optimised kernel , i use pm-utils , should i switch to hibernate-script although it has the same problem...One thing that is peculiar about its hibernation is after "...kernel and process data.." the caps lock led blinks , but when i power down and reboot , it resumes successfully , which implies that hibernate works fine , only that it doesnt shutdown/halt after hibernation and throws led blinks at me , Besides, , the repo has pf8 whereas aur has pf10 , should i continue using the repo , or change to AUR , it ll be nice if you do keep updating the repo though...
Do you also put the Xorg ISO_Scheduling patch , it seems to improve the responsiveness of xserver? If its not there in the default PKGBUID , i ll try to manually patch it....
I try to upload the latest packages as fast as I can, but real life has to take precedence. I'm compiling the i686-pf10 right now, the 64-bit versions should be up by tomorrow evening.
Since time is a constraint, I try to keep the included patches few and maintainable. The -pf patchset itself anyway contains almost everything a real-time fan could ask. And, to be honest, if an i7 isn't snappy enough on -pf alone, then no tweak can help you ;-)

The pf-kernel seems the best to me currently but hibernation is nonetheless an important feature.
since i can resume successfully after manually powering down , can i consider this a safe practice or is there a way i can prevent the kernel panic at the end of hibernation... I remember someone else in this post also had problem during shutdown , led blinking similar to me ...

nous · 2010-09-29 20:01:27

pankajmore wrote:

t to me currently but hibernation is nonetheless an important feature.
since i can resume successfully after manually powering down , can i consider this a safe practice or is there a way i can prevent the kernel panic at the end of hibernation...

Since you can resume successfully, it's safe. It just never reaches the ACPI poweroff call. Do you still have problems with pf10 from the AUR?

Last edited by nous (2010-09-29 20:03:32)

pankajmore · 2010-09-29 20:10:39

nous wrote:

pankajmore wrote:
t to me currently but hibernation is nonetheless an important feature.
since i can resume successfully after manually powering down , can i consider this a safe practice or is there a way i can prevent the kernel panic at the end of hibernation...
Since you can resume successfully, it's safe. It just never reaches the ACPI poweroff call. Do you still have problems with pf10 from the AUR?

when i updated to pf10 , i could hibernate with pm-hibernate for the first tym, after that , subsequent pm-hibernate's lead to the same led blink issue...:( Is there anyway to force poweroff after hibernation...

nous · 2010-10-01 20:35:55

pankajmore wrote:

Is there anyway to force poweroff after hibernation...

acpi=force

pankajmore · 2010-10-01 20:49:03

nous wrote:

pankajmore wrote:
Is there anyway to force poweroff after hibernation...
acpi=force

sadly, this option doesn't work , i still have caps lock led blinking issue after hibernate , any other suggestions please...

nous · 2010-10-11 06:48:12

I've included the experimental fork depth patch from Con Kolivas' blog in the latest PKGBUILD. It worked wonders for me, so please feel free to test. The binaries of kernel26-pf in the repos are still vanilla -pf, just in case.

stqn · 2010-10-14 12:03:06

Thanks for this package. I've been using kernel26-pf11 with the fork thing patch for the last two or three days without problem. I can't say I have noticed any particular improvement against my usual unpatched 2.6.34 build, but I haven't compared either.

I had noticed improvements with games and video playback with other BFS-patched kernels in the recent past, but also problems (crashes); but no crash so far with pf11.

However, something is puzzling me. Two things in fact:
- I have swappiness set to 0, but swap is still being used (211MB currently, and "free -m" tells me I have 522MB of free RAM.)
- What is LinuxIMQ ? I can't understand anything on its home page. (I don't know about SquashFS and AUFS either, but I'm assuming they are filesystems and not in use by default, so I haven't had a look at them.)

Last edited by stqn (2010-10-14 16:11:33)

nous · 2010-10-14 15:13:24

stqn wrote:

However, something is puzzling me. Two things in fact:
- I have swapiness set to 0, but swap is still being used (211MB currently, and "free -m" tells me I have 522MB of free RAM.)
- What is LinuxIMQ ? I can't understand anything on its home page. (I don't know about SquashFS and AUFS either, but I'm assuming they are filesystems and not in use by default, so I haven't had a look at them.)

On my laptop, the swappiness value seems to be observed:

orion:[nous]:~% cat /proc/sys/vm/swappiness 
0
orion:[nous]:~% free -m
             total       used       free     shared    buffers     cached
Mem:          3960       3846        113          0        147        854
-/+ buffers/cache:       2844       1116
Swap:         2478          0       2478

Try adding vm.swappiness=0 in /etc/sysctl.conf (but I guess you already know that).

IMQ is a traffic shaping device. I don't use it and it's off by default. Chances are, if one doesn't know what ingress shaping is (like me) one won't need it.

stqn · 2010-10-15 13:41:13

nous wrote:

Try adding vm.swappiness=0 in /etc/sysctl.conf (but I guess you already know that).

I was setting it to 0 in my rc.local, so I also changed sysctl.conf, but this doesn't seem to change anything. After boot swap usage is 0, but a couple of hours later I get:

.............total       used       free     shared    buffers     cached
Mem:          1003        924         79          0        268        359
-/+ buffers/cache:        296        707
Swap:          588         44        543

Anyway, one of the CK patches sets swappiness to 0 by default.

I'll have to check if this happens also with another kernel... I also wonder what can be using so much swap space, because when I sum the RAM usage of my running apps, it doesn't add up.

Last edited by stqn (2010-10-15 13:44:27)

nous · 2010-10-15 15:08:07

stqn wrote:

Anyway, one of the CK patches sets swappiness to 0 by default.
I'll have to check if this happens also with another kernel... I also wonder what can be using so much swap space, because when I sum the RAM usage of my running apps, it doesn't add up.

Don't get confused: even with a swappiness value of 0, the kernel WILL swap out to disk if your physical RAM is exhausted, starting with the less frequently used segments. In your case is far more likely to happen, since you only have 1GB installed. Even if free(1) shows available memory as buffers, there's a safety threshold the kernel doesn't cross and/or there might be a moment the kernel had to swap out because the buffers were zeroed.

arch0r · 2010-10-16 17:21:18

thx a lot for this kernel package, but i can't hardly notice any great performance improvements on my system.
is there any benchmark which shows the differences of a vanilla and the pf patched kernel? or can i run a benchmark to prove that there's no placebo effect?

stqn · 2010-10-16 17:32:13

stqn wrote:

I'll have to check if this happens also with another kernel...

Ok, I'm running my previous unpatched 2.6.34 kernel (only non standard thing is the speedstep-centrino module), and running the exact same programs as before, after three hours, swap is still unused.

.............total       used       free     shared    buffers     cached
Mem:          1010        992         17          0        174        292
-/+ buffers/cache:        525        484
Swap:          588          0        588

So there is something that forces swap usage in the pf11 kernel...

Last edited by stqn (2010-10-16 17:35:52)

nous · 2010-10-16 20:32:14

stqn wrote:

stqn wrote:
I'll have to check if this happens also with another kernel...
Ok, I'm running my previous unpatched 2.6.34 kernel (only non standard thing is the speedstep-centrino module), and running the exact same programs as before, after three hours, swap is still unused.
So there is something that forces swap usage in the pf11 kernel...

You're probably right. On my desktop, which only has 2GB of RAM, a few MB are swapped out. Could you also test with the latest 2.6.35-ARCH, to eliminate the version bump doubt?

nous · 2010-10-16 21:33:50

arch0r wrote:

thx a lot for this kernel package, but i can't hardly notice any great performance improvements on my system.
is there any benchmark which shows the differences of a vanilla and the pf patched kernel? or can i run a benchmark to prove that there's no placebo effect?

There's interbench and the phoronix test suite, for starters, but I haven't tried them because I didn't need to. My major woe in the past (i.e. before BFS) was audio and video frame skipping under heavy disk activity and/or increased load, which I found unacceptable on my dual-core laptop. After switching to BFS I never had a single audio skip since. That is my personal benchmark and it's no placebo. Also, when I posted my enthusiastic experience in this forum, I actually got bashed for not being a good linux user because I didn't provide feedback to the CFQ developers and I didn't tweak the default CFQ behavior.

The default scheduler nowdays is not as bad as it used to be, but that's thanks to CK's scheduler which drew a lot of attention to the desktop responsiveness. BFS though is still better for desktop use and if you take a look at the AUR you'll find that the most popular kernel26 packages incorporate the BFS patch. In fact, I'm thinking of abandoning kernel26-pf since those kernel26s sport the features I use most (BFS and TuxOnIce).

stqn · 2010-10-18 16:21:00

nous wrote:

On my desktop, which only has 2GB of RAM, a few MB are swapped out. Could you also test with the latest 2.6.35-ARCH, to eliminate the version bump doubt?

Yep, I used the official 2.6.35.7 ARCH kernel yesterday for more than 5 hours and swap wasn't used.

Today with pf11, after only 1 h and 25 min, I have 25 MB of swap in use, using the same programs and doing nothing special (no compilation or image manipulation or 3D rendering...) (I tried again because I upgraded my system yesterday with the new xorg etc.)

It doesn't seem to cause any problem or slowdown though... But it's strange.

Cdh · 2010-11-15 22:47:20

Can we get CONFIG_TIMER_STATS enabled for powertop please? Or is it bad for performance or anything?

nous · 2010-11-16 11:56:06

Cdh wrote:

Can we get CONFIG_TIMER_STATS enabled for powertop please? Or is it bad for performance or anything?

Done (for i686 where it wasn't). I'm not updating the binary repo though until -pf2.

Cdh · 2010-11-16 12:47:39

WAIT! Powertops says, that inoitfy is missing too:

Suggestion: Enable the CONFIG_INOTIFY kernel configuration option.
This option allows programs to wait for changes in files and directories
instead of having to poll for these changes

bangkok_manouel · 2010-11-16 12:58:01

you should make your own config (not that I disagree with you, not enabling inotify is a surprising move)

bluepumpkin · 2010-11-16 15:04:08

Cdh wrote:

WAIT! Powertops says, that inoitfy is missing too:
Suggestion: Enable the CONFIG_INOTIFY kernel configuration option.
This option allows programs to wait for changes in files and directories
instead of having to poll for these changes

I think it is a bug in powertop. I'm running kernel26-ck and specifically checked that it WAS enabled because of this powertop message. It is, in fact, enabled and working (tested with inotail).

Last edited by bluepumpkin (2010-11-16 15:04:25)

Arch Linux

#76 2010-09-28 19:14:26

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#77 2010-09-29 16:15:59

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#78 2010-09-29 16:37:33

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#79 2010-09-29 17:32:15

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#80 2010-09-29 17:37:27

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#81 2010-09-29 18:19:40

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#82 2010-09-29 20:01:27

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#83 2010-09-29 20:10:39

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#84 2010-10-01 20:35:55

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#85 2010-10-01 20:49:03

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#86 2010-10-11 06:48:12

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#87 2010-10-14 12:03:06

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#88 2010-10-14 15:13:24

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#89 2010-10-15 13:41:13

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#90 2010-10-15 15:08:07

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#91 2010-10-16 17:21:18

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#92 2010-10-16 17:32:13

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#93 2010-10-16 20:32:14

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#94 2010-10-16 21:33:50

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#95 2010-10-18 16:21:00

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#96 2010-11-15 22:47:20

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#97 2010-11-16 11:56:06

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#98 2010-11-16 12:47:39

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#99 2010-11-16 12:58:01

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#100 2010-11-16 15:04:08

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

Board footer