You are not logged in.

#1 2007-07-15 05:00:01

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Anyone else getting reduced performance with kernel 2.6.22?

Since upgrading (in 32 bit Arch), I've noticed that applications take a noticably longer time to load for the first time, with the hard drive light indicating more activity. Is it possible that some of those CFQ modifications in the most recent kernel can harm read performance under certain circumstances?

Offline

#2 2007-07-15 07:03:21

kishd
Member
Registered: 2006-06-14
Posts: 401

Re: Anyone else getting reduced performance with kernel 2.6.22?

I have an Asus mb with 4gb ram. I used the stock kernel and noticed that the system seemed a little slower than before. I had to recompile the kernel to enable PAE to get the kernel to see all my ram. I noticed that the new memory option SLUB unique allocator was an option. I enabled it and now the system seems (subjectively) noticeably faster. I read a bit about this option at kernel newbies but still do not understand it fullly.


---for there is nothing either good or bad, but only thinking makes it so....
Hamlet, W Shakespeare

Offline

#3 2007-07-15 07:26:29

onearm
Member
From: Anywhere but here
Registered: 2006-07-06
Posts: 359
Website

Re: Anyone else getting reduced performance with kernel 2.6.22?

I noticed the slowing of the system too, especially regarding to disk access. For example mounting my crypted partition with the 2.6.22 takes at least 2 seconds while with 2.6.21 it didn't take more than 1.


To get something done, a committee should consist of no more than three persons, two of them absent.
--
My Github

Offline

#4 2007-07-15 08:16:45

Acid7711
Member
From: Chicago, IL
Registered: 2006-08-18
Posts: 300
Website

Re: Anyone else getting reduced performance with kernel 2.6.22?

Viper patcheset ftw!  This in itself for me seems faster than the base kernel, although I haven't tried that in some time.

Last edited by Acid7711 (2007-07-15 08:17:57)

Offline

#5 2007-07-15 09:36:42

Cotton
Member
From: Cornwall, UK
Registered: 2004-09-17
Posts: 568

Re: Anyone else getting reduced performance with kernel 2.6.22?

Following advice elsewhere on these forums, I enabled the SLUB allocator and recompiled. 2.6.22.1 seems the fastest yet.  Even gives viper a run for its money.

Offline

#6 2007-07-15 13:33:07

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

I'm thinking of switching the scheduler to anticipatory, to see if that helps the speed.

Re SLUB: if that's compiled in, can the option to use SLAB or SLUB be specified at boot time? Or are they mutually exclusive? If SLUB/SLAB can be specified on boot it might be good to request that the Testing kernel be compiled with SLUB support.

Offline

#7 2007-07-15 14:36:46

sula
Member
Registered: 2005-08-07
Posts: 93

Re: Anyone else getting reduced performance with kernel 2.6.22?

So maybe kernel in repositories should be recompiled with SLUB and PAE before it hits current.
I notice slowdowns with it too.

Offline

#8 2007-07-15 15:24:47

kishd
Member
Registered: 2006-06-14
Posts: 401

Re: Anyone else getting reduced performance with kernel 2.6.22?

SLUB seems to give a performance improvement. PAE definitely does not increase performance. In fact it decreases performance but is necessary for the kernel to see > 4GB ram

Here is an extract from kernelnewbies

New Slab allocator: SLUB

(Recommended article: "The SLUB allocator")

The slab allocator is a object-caching kernel memory allocator used for dealing with "objects that are frequently allocated and freed" (see the "slab allocator" paper from Jeff Bonwick). It is a critical piece of the innards of the memory management subsystem, and a critical piece to get good performance. The Linux slab allocator works quite well for pretty much everybody; however some people (SGI) has found its current design inefficient in some cases. For example, in 1K nodes/processors configurations, several GB of memory are wasted only in object queues, not counting the objects themselves. The memory management quickly becomes too complex when adding features like proper NUMA policy support.

As result, a new slab allocator called "SLUB" has been developed by Christoph Lameter from SGI, to solve those and other problems. Its design is simpler, but it also addresses some problems that can result in better performance in some cases and more efficient memory usage (see the full design notes in this commit link). It also has better debug capabilities. There's a slabinfo userspace tool that you can find in Documentation/vm/slabinfo.c.

Its aim is to transparently replace slab, but in 2.6 this new slab allocator is optional and not enabled by default. You can enable it at compile time (making it the third option along with SLOB, the embedded-oriented slab allocator). SLUB has been tested for some time and it's solid enough to try it on your systems, but due to the importance of this part of the kernel, it won't completely replace the current slab allocator until more exposure and testing has been done, hence it's not recommended to use it in production systems. Testing reports, specially regressions, are greatly appreciated.


---for there is nothing either good or bad, but only thinking makes it so....
Hamlet, W Shakespeare

Offline

#9 2007-07-15 17:59:06

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

So... No boot option? Is it incompatible with SLAB, or can you enable it and disable SLAB at boot (or the opposite)?

Offline

#10 2007-07-15 19:08:47

kishd
Member
Registered: 2006-06-14
Posts: 401

Re: Anyone else getting reduced performance with kernel 2.6.22?

It's either SLOB, SLAB or SLUB enabled only at compile time.


---for there is nothing either good or bad, but only thinking makes it so....
Hamlet, W Shakespeare

Offline

#11 2007-07-15 20:05:48

brain0
Developer
From: Aachen - Germany
Registered: 2005-01-03
Posts: 1,382

Re: Anyone else getting reduced performance with kernel 2.6.22?

As far as I understand this, SLUB should only improve performance with many CPUs. All previous kernels only had SLAB, so I don't understand why performance should be worse now.

We could enable SLUB, but we won't enable PAE. If possible, you should use a 64 bit system with more than 3GB RAM.

Offline

#12 2007-07-15 22:09:58

sula
Member
Registered: 2005-08-07
Posts: 93

Re: Anyone else getting reduced performance with kernel 2.6.22?

I'll vote for SLUB big_smile

I would also want PAE but i understand reasons why it would not be good to turn it on in arch vanilla...

Offline

#13 2007-07-16 09:55:49

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

What I'm wondering is where the extra HDD activity is coming from. I'm starting to think that 2.6.22 handles swap differently from previous kernels...

Offline

#14 2007-07-16 11:14:20

kishd
Member
Registered: 2006-06-14
Posts: 401

Re: Anyone else getting reduced performance with kernel 2.6.22?

After some googleing found this thread

http://readlist.com/lists/gentoo.org/ge … 94155.html

Following the advice in the thread I disabled the following in the kernel config

Generic ATA Support
Generic PCI/IDE chipset support
Generic/Default IDE chipset support

My file access etc are much faster and now the occassional interuoptions I got with amarok during heavy disk activity stopped.

PS I also found ata_generic in my mkinitcpio.conf so this module was loaded before any other module. This may be the case for others experiencing the slower disk access.

Last edited by kishd (2007-07-16 11:54:44)


---for there is nothing either good or bad, but only thinking makes it so....
Hamlet, W Shakespeare

Offline

#15 2007-07-18 05:11:49

PenguinFlavored
Member
From: Chicago, IL, USA
Registered: 2006-06-06
Posts: 66
Website

Re: Anyone else getting reduced performance with kernel 2.6.22?

I just compiled and installed linux 2.6.22.1-suspend2 today, and I've been noticing random freezes every so often. Everything would just stop, no mouse, no gkrellm movement, and after a few seconds (got up to a minute once). I haven't had time to investigate anything yet, but nothing appears in dmesg. I don't know if this is because of the latest madwifi drivers, the suspend2 patch, or linux itself. Is anyone else getting these freezes?

Offline

#16 2007-07-18 17:29:41

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

kishd wrote:

After some googleing found this thread

http://readlist.com/lists/gentoo.org/ge … 94155.html

Following the advice in the thread I disabled the following in the kernel config

Generic ATA Support
Generic PCI/IDE chipset support
Generic/Default IDE chipset support

My file access etc are much faster and now the occassional interuoptions I got with amarok during heavy disk activity stopped.

PS I also found ata_generic in my mkinitcpio.conf so this module was loaded before any other module. This may be the case for others experiencing the slower disk access.

Hmm, I had ata_generic not loaded when I noticed the performance problems. I tried putting it in but it didn't make any difference either way. It hasn't made a difference either way in previous kernels either.

Edit: Hmm, I retract the above. Since adding ata_generic to my initcpio image, I've found that my system freezes for a few seconds a few times an hour. There's definitely something fishy going on here, hopefully the next kernel update will fix it...

Last edited by Gullible Jones (2007-07-18 18:46:30)

Offline

#17 2007-07-18 17:31:12

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

PenguinFlavored: are you using any sort of CPU frequency control? If so, try turning it off.

Offline

#18 2007-07-21 21:01:22

PenguinFlavored
Member
From: Chicago, IL, USA
Registered: 2006-06-06
Posts: 66
Website

Re: Anyone else getting reduced performance with kernel 2.6.22?

Gullible Jones wrote:

PenguinFlavored: are you using any sort of CPU frequency control? If so, try turning it off.

I am using frequency scaling, turning it off didn't help. After googling around I've found out that it's related to the clocksource. Using tsc freezes it upon frequency scaling. However I tried setting the cpu frequency back and forth from performance to powersave, making sure the cpu MHz changes, which it does. That doesn't change it. Adding clocksource=acpi-pm to my kernel command line seems to solve the problem, so far I haven't been getting these system freezes.

Looking at my dmesg more carefully I get:

Clocksource tsc unstable (delta = -80357688 ns)

This happens with clocksource=acpi-pm and without.

Thanks for the tip Gullible Jones smile

Offline

#19 2007-07-21 21:08:10

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

You're welcome. I recently got something like you did when I was trying to get cpufreq working with my laptop, and found that I would freeze about 15 seconds after I finished booting unless I specified the governor in powersave's config - it was supposed to default to ondemand, but apparently it didn't and something screwed up.

(I don't bother with cpufreq now, of course, since Celeron 520s can only use p4_clockmod, and that kills performance and doesn't save me any power.)

Offline

#20 2007-07-23 07:44:21

steve_v
Member
Registered: 2006-02-11
Posts: 80

Re: Anyone else getting reduced performance with kernel 2.6.22?

Same sort of issues here, any heavy disk io (ie. pacman) completley cripples desktop responsiveness.
Only noticed after upgrade to 2.6.22

Offline

#21 2007-07-23 16:35:19

skottish
Forum Fellow
From: Here
Registered: 2006-06-16
Posts: 7,942

Re: Anyone else getting reduced performance with kernel 2.6.22?

steve_v wrote:

Same sort of issues here, any heavy disk io (ie. pacman) completley cripples desktop responsiveness.
Only noticed after upgrade to 2.6.22

It's been like this with the CK kernel for a while, and I'm not even using 2.6.22 yet.

Offline

#22 2007-07-24 07:51:34

brain0
Developer
From: Aachen - Germany
Registered: 2005-01-03
Posts: 1,382

Re: Anyone else getting reduced performance with kernel 2.6.22?

Anyone having these problems, could you try removing "ata_generic" and/or "generic" from your MODULES= line in mkinitcpio.conf, then run mkinitcpio -p kernel26? It has been suggested above as a solution to the problem above.

Offline

#23 2007-07-24 09:19:14

PDExperiment626
Member
From: Australia
Registered: 2007-04-02
Posts: 66

Re: Anyone else getting reduced performance with kernel 2.6.22?

My problem is when I remove the IDE_generic driver from being built in the kernel I no longer can use the DVD burner on my laptop. I've tried to use ide-cd with the piix ide drivers to no avail. The only way I've been able to get my ide burner to work under linux is to build ide-generic and ide-cd into the kernel (the the burner doesn't work if they are compiled as modules). I load ata_piix as I have a SATA HD. If I compile ide_generic into my 2.6.22 kernel the system doesn't boot (kicks me out to an emergency prompt). Any help is greatly appreciated.


... and for a time, it was good...

Offline

#24 2007-07-24 16:31:35

brain0
Developer
From: Aachen - Germany
Registered: 2005-01-03
Posts: 1,382

Re: Anyone else getting reduced performance with kernel 2.6.22?

You seem to use ata_piix, which takes control of all Intel ide AND sata chips when loaded (and it is, by default, loaded before piix). That means, your cd writer will use the scsi/ata subsystem. My best guess is that you didn't add "SCSI generic support" to your custom kernel, which is required in that case.

Offline

#25 2007-08-15 00:31:24

Gullible Jones
Member
Registered: 2004-12-29
Posts: 4,863

Re: Anyone else getting reduced performance with kernel 2.6.22?

Sorry for the resurrection but I think I know what the problem might be... I discovered that adding highres=off nohz=off to GRUB's kernel line solved my suspend problems, as indicated in other threads, and it also seems to improve my machine's performance. I don't know much about the dynamic Hz stuff that's being introduced to the kernel, but I wonder if the way it's implemented right now could reduce performance in addition to screwing up software suspend (if it's what's actually responsible).

Offline

Board footer

Powered by FluxBB