You are not logged in.

#101 2010-09-29 06:05:09

Ranguvar
Member
Registered: 2008-08-12
Posts: 2,549

Re: (Solved) Kernel 2.6.35 I/O congestion

I've had issues like those being described for a long time, and 2.6.35 didn't change anything for me.
A while ago I had messed with BFS a tad without noticing an improvement, but recently I custom-compiled a kernel with CK's patchset as well as the BFQ I/O scheduler, and I am blown away.
Before, even a nice -n19 on a compile would still cause serious FPS problems in games, but now I can't even tell anything else is running.

Sorry that this post doesn't offer a solution to the stock kernel issues, and I know CK+BFQ doesn't fix them for everyone, but I highly recommend giving it a try smile

Last edited by Ranguvar (2010-09-29 06:05:32)

Offline

#102 2010-09-29 11:08:40

archman-cro
Member
From: Croatia
Registered: 2010-04-04
Posts: 943
Website

Re: (Solved) Kernel 2.6.35 I/O congestion

broch wrote:

unfortunately two latest bfs aren't great (latest 350 for obvious reasons because this is major scheduler overhaul), so to kkep in nice either run bfs on 2.6.34.x or 323 on 2.6.35.x

I've been using 330, cause I didn't even see the 350 out. Do you really experience problems with 350 or you just read it somewhere? I might test the new release...

Offline

#103 2010-09-29 11:52:15

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: (Solved) Kernel 2.6.35 I/O congestion

Keep it ontopic guys, there are other threads for BFS. smile


ᶘ ᵒᴥᵒᶅ

Offline

#104 2010-09-29 13:22:06

broch
Banned
From: L.A. California
Registered: 2006-11-13
Posts: 975

Re: (Solved) Kernel 2.6.35 I/O congestion

sorry about bfs
I noticed though (cfs kernel), that not only powertop but also infrequent keyboard activity decreases idle (hope that meaning is clear) load. So issue may be somehow related to idle cpu scheduling. This is additional to the issue with cfs and high load.

Offline

#105 2010-09-30 08:11:08

SleepyFloyd
Member
Registered: 2008-06-11
Posts: 91

Re: (Solved) Kernel 2.6.35 I/O congestion

Did 2.6.35.7 fix the problem?

(too lazy to play the upgrade&downgrade game myself...)

Offline

#106 2010-09-30 11:32:53

archman-cro
Member
From: Croatia
Registered: 2010-04-04
Posts: 943
Website

Re: (Solved) Kernel 2.6.35 I/O congestion

I didn't try, but I've been experiencing some lag on .6 today. I'll upgrade today.

Offline

#107 2010-09-30 13:44:14

broch
Banned
From: L.A. California
Registered: 2006-11-13
Posts: 975

Re: (Solved) Kernel 2.6.35 I/O congestion

uptime (2.6.35.6 kernel-bfs)
06:42:04 up 7 days, 45 min,  1 user,  load average: 0.05, 0.19, 0.35

Offline

#108 2010-09-30 13:46:24

karol
Archivist
Registered: 2009-05-06
Posts: 25,440

Re: (Solved) Kernel 2.6.35 I/O congestion

broch wrote:

uptime (2.6.35.6 kernel-bfs)
06:42:04 up 7 days, 45 min,  1 user,  load average: 0.05, 0.19, 0.35

Could you post the stats for the stock kernel?

Offline

#109 2010-09-30 14:45:50

broch
Banned
From: L.A. California
Registered: 2006-11-13
Posts: 975

Re: (Solved) Kernel 2.6.35 I/O congestion

I don't have stock kernel installed (I removed Arch kernel just after installation, more than 3-yrs ago on this machine and on the netbook just after OS installation 6mos ago). The only comparison I did was between two kernels with two CPU schedulers as only difference that were run under exactly the same conditions on the netbook Toshiba nb305-n410.

Tested kernels were vanilla with the following patches added:
laptop and netbook:
bfs 323

cat include/linux/sched.h | grep -i kolivas
printk(KERN_INFO"BFS CPU scheduler v0.323 by Con Kolivas.\n");

http://ck.kolivas.org/patches/bfs/

netbook only
patch that restores LCD brightness control for the netbook (it basically removes some recent ACPI patches for i915)
https://bugzilla.kernel.org/attachment.cgi?id=27236


All kernels use the same I/O scheduler noop (best for xfs).

load averages are reaching low levels very fast (for example after compiling broken arch xulrunner)

Last edited by broch (2010-09-30 14:49:38)

Offline

#110 2010-09-30 15:04:23

karol
Archivist
Registered: 2009-05-06
Posts: 25,440

Re: (Solved) Kernel 2.6.35 I/O congestion

I think I need to be more blunt.
<whistles innocently>
https://bbs.archlinux.org/viewtopic.php … 35#p832935

Offline

#111 2010-09-30 15:46:07

broch
Banned
From: L.A. California
Registered: 2006-11-13
Posts: 975

Re: (Solved) Kernel 2.6.35 I/O congestion

jesus.
leitmotif referrent to the posts that "cropped" after I compared two kernels (because thread is not about praising cpu or I/O scheduler). It is not about BFS but finding reason what is wrong with elevated load.
to be blunt: if cpu scheduler is at fault then excluding from discussion this does not make any sense. and this is not about switching schedulers but finding the culprit.

this is second time in this thread where you are posting off topic. Maybe would be better if you either stop posting, or add something more constructive.

"nohz=off highres=off" helps a lot.

this is not sane solution for anyone with laptop because this will drain battery like crazy.

Last edited by broch (2010-09-30 15:49:33)

Offline

#112 2010-09-30 16:30:12

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: (Solved) Kernel 2.6.35 I/O congestion

Ok, what i would like to hear is if anyone can confirm my bisection results. You'll have to use the git-kernel for this and try out two commits:

good: 66f41d4c5c8a5deed66fdcc84509376c9a0bf9d8
bad:  e7858f52a5cb868289a72264534a1f05f3340c6c

If this gets confirmed, then we are very close to both the cause and solution.


ᶘ ᵒᴥᵒᶅ

Offline

#113 2010-10-01 18:03:09

bluepumpkin
Member
Registered: 2009-08-28
Posts: 58

Re: (Solved) Kernel 2.6.35 I/O congestion

2.6.35.7 fixed it for me:

load average: 0.00, 0.00, 0.00

EDIT: Well, a few seconds after I posted, my load shot up to ~.45 for no obvious reason. I guess it's still acting a bit strange.

Last edited by bluepumpkin (2010-10-01 18:05:24)

Offline

#114 2010-10-01 21:27:41

archman-cro
Member
From: Croatia
Registered: 2010-04-04
Posts: 943
Website

Re: (Solved) Kernel 2.6.35 I/O congestion

Pretty low for me on .7 for now (on a normal usage (browsing, several apps open):

23:26:40 up 1 day,  7:49,  1 user,  load average: 0.09, 0.32, 0.30

Offline

#115 2010-10-03 05:14:31

Ranguvar
Member
Registered: 2008-08-12
Posts: 2,549

Re: (Solved) Kernel 2.6.35 I/O congestion

2.6.35.7 just fixes Xen, which was broken by a tiny typo, so unless you're comparing from <2.6.35.6, it's a placebo.

Offline

#116 2010-10-07 20:49:18

Texas
Member
From: Dallas, Texas
Registered: 2010-09-10
Posts: 131

Re: (Solved) Kernel 2.6.35 I/O congestion

How does one downgrade to 2.6.34?  When I 'pacman -Ss kernel', I only see 2.6.35 and 2.6.32 (lts).

Last edited by Texas (2010-10-07 21:12:03)

Offline

#117 2010-10-07 20:57:15

karol
Archivist
Registered: 2009-05-06
Posts: 25,440

Re: (Solved) Kernel 2.6.35 I/O congestion

Texas wrote:

How does one downgrade to 2.6.34?  When I 'pacman -Ss kerne', I only see 2.6.35 and 2.6.32 (lts).

The wiki isn't perfect, but that doesn't mean you should ignore it http://wiki.archlinux.org/index.php/Downgrade#ARM

Offline

#118 2010-10-07 21:12:34

KimTjik
Member
From: Sweden
Registered: 2007-08-22
Posts: 715

Re: (Solved) Kernel 2.6.35 I/O congestion

Not a solution, but I noticed the same issue on a computer at work, running a typical Intel P4 CPU. Load is also very high:

load average: 1.49, 1.57, 1.24

Offline

#119 2010-10-07 21:18:46

Texas
Member
From: Dallas, Texas
Registered: 2010-09-10
Posts: 131

Re: (Solved) Kernel 2.6.35 I/O congestion

karol wrote:

I think I need to be more blunt.
<whistles innocently>
https://bbs.archlinux.org/viewtopic.php … 35#p832935

Sorry... I'm still fairly new to Arch and only knew of pacman and AUR.  I'll try this tonight.

Offline

#120 2010-10-09 11:04:55

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: (Solved) Kernel 2.6.35 I/O congestion

FYI: i tried kernel26-pf (with BFS) and it doesn't really seem to improve the situation.


ᶘ ᵒᴥᵒᶅ

Offline

#121 2010-10-22 10:53:35

Damnshock
Member
From: Barcelona
Registered: 2006-09-13
Posts: 414

Re: (Solved) Kernel 2.6.35 I/O congestion

Schnouki wrote:

"nohz=off" and "highress=off" really really help here.

To me it's like using a different computer.

Thanks a lot!


My blog: blog.marcdeop.com
Jabber ID: damnshock@jabber.org

Offline

#122 2010-10-22 12:55:51

KimTjik
Member
From: Sweden
Registered: 2007-08-22
Posts: 715

Re: (Solved) Kernel 2.6.35 I/O congestion

KimTjik wrote:

Not a solution, but I noticed the same issue on a computer at work, running a typical Intel P4 CPU. Load is also very high:

load average: 1.49, 1.57, 1.24

I'm quoting myself, just to clarify that the issue I reported probably has nothing to do with a kernel issue. Further investigation show that there's reason to believe that it's caused by a hard drive failure. Drive will be cloned and swapped.

Offline

#123 2010-12-05 18:15:49

karol
Archivist
Registered: 2009-05-06
Posts: 25,440

Re: (Solved) Kernel 2.6.35 I/O congestion

I'm using kernel26 2.6.36.1-3 atm and I still get

[karol@black ~]$ uptime
 19:14:27 up  5:11,  1 user,  load average: 0.86, 0.85, 0.91

without any apparent reason.

Offline

#124 2010-12-05 18:33:32

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: (Solved) Kernel 2.6.35 I/O congestion

karol wrote:

I'm using kernel26 2.6.36.1-3 atm and I still get

[karol@black ~]$ uptime
 19:14:27 up  5:11,  1 user,  load average: 0.86, 0.85, 0.91

without any apparent reason.

A kernel patch has been released, but will probably not be there before 2.6.37 since it's a core/scheduler change


ᶘ ᵒᴥᵒᶅ

Offline

#125 2010-12-05 20:01:03

combuster
Member
From: Serbia
Registered: 2008-09-30
Posts: 711
Website

Re: (Solved) Kernel 2.6.35 I/O congestion

Can you provide a link to the LKML/bugzilla/kernel.org thread/patch ? Thanks.

Offline

Board footer

Powered by FluxBB