Skip to contentSkip to conversation

2.6.1 slower than 2.4, smp/scsi/sw-raid/reiserfs

DISCUSSION Samium Gromoff 4 participants 6 messages Feb 3, 2004

lore
Share & tools
lore
#0 lore raw
Quoted 8 lines
Quoted 5 lines

The machine is a dual P3 450MHz, 512MB, aic7xxx, 2 disk RAID-0 and
ReiserFS. It's a few years old and has always run Linux, most recently 2.4.24. I decided to try 2.6.1 and the performance is disappointing.

2.6 has a few performance problems under heavy pageout at present. Nick Piggin has some patches which largely fix it up.

I`m sorry, but this is misguiding. 2.6 does not have a few performance problems under heavy pageout.

It`s more like _systematical_ _performance_ _degradation_ increasing with the pageout rate. The more the box pages out the more 2.6 lags behind 2.4.

What i`m trying to say is that even light paging is affected. And light paging is warranted when you run, say, KDE on 128M ram.

Go measure the X desktop startup time on a 48M/64M boxen--even light paging causes 2.6 to be just sloower. Also the vm thrashing point is much much earlier.

Ask Roger Luethi for details.

regards, Samium Gromoff

#1 lore raw
Quoted 7 lines · Samium Gromoff
Quoted 7 lines
Quoted 6 lines

The machine is a dual P3 450MHz, 512MB, aic7xxx, 2 disk RAID-0 and
ReiserFS. It's a few years old and has always run Linux, most recently 2.4.24. I decided to try 2.6.1 and the performance is disappointing.

Quoted 6 lines
Quoted 2 lines

2.6 has a few performance problems under heavy pageout at present. Nick Piggin has some patches which largely fix it up.

I`m sorry, but this is misguiding. 2.6 does not have a few performance problems under heavy pageout.

As you have frequently and somewhat redundantly reminded us.

Perhaps you could test Nick's patches. They are at

http://www.kerneltrap.org/~npiggin/vm/

Against 2.6.2-rc2-mm2. First revert vm-rss-limit-enforcement.patch, then apply those three.

#2 lore raw

At Mon, 2 Feb 2004 23:07:45 -0800,

Quoted 9 lines · Andrew Morton
Quoted 7 lines · Samium Gromoff
Quoted 7 lines
Quoted 6 lines

The machine is a dual P3 450MHz, 512MB, aic7xxx, 2 disk RAID-0 and
ReiserFS. It's a few years old and has always run Linux, most recently 2.4.24. I decided to try 2.6.1 and the performance is disappointing.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

2.6 has a few performance problems under heavy pageout at present. Nick Piggin has some patches which largely fix it up.

I`m sorry, but this is misguiding. 2.6 does not have a few performance problems under heavy pageout.

As you have frequently and somewhat redundantly reminded us.

Right. I`m rather emotional about this...

Kind of hard seeing the all starry and shiny 2.6 going down the toilet on my little server with 16M RAM.

Quoted 6 lines

Perhaps you could test Nick's patches. They are at

http://www.kerneltrap.org/~npiggin/vm/

Against 2.6.2-rc2-mm2. First revert vm-rss-limit-enforcement.patch, then apply those three.

Hmmm, id prefer plain 2.6, but ill try it anyway.

regards, Samium Gromoff

#3 lore raw
Quoted 34 lines · Samium Gromoff

At Mon, 2 Feb 2004 23:07:45 -0800,

Quoted 12 lines · Andrew Morton
Quoted 9 lines · Samium Gromoff
Quoted 9 lines
Quoted 5 lines

The machine is a dual P3 450MHz, 512MB, aic7xxx, 2 disk RAID-0 and ReiserFS. It's a few years old and has always run Linux, most recently 2.4.24. I decided to try 2.6.1 and the performance is disappointing.

2.6 has a few performance problems under heavy pageout at present. Nick Piggin has some patches which largely fix it up.

Quoted 8 lines
Quoted 5 lines

I`m sorry, but this is misguiding. 2.6 does not have a few performance problems under heavy pageout.

As you have frequently and somewhat redundantly reminded us.

Right. I`m rather emotional about this...

Kind of hard seeing the all starry and shiny 2.6 going down the toilet on my little server with 16M RAM.

Quoted 2 lines

Perhaps you could test Nick's patches. They are at

Quoted 11 lines
Quoted 6 lines

http://www.kerneltrap.org/~npiggin/vm/

Against 2.6.2-rc2-mm2. First revert vm-rss-limit-enforcement.patch, then apply those three.

Hmmm, id prefer plain 2.6, but ill try it anyway.

It should go against plain 2.6 with luck.

#4 lore raw
Quoted 8 lines · Nick Piggin
Quoted 7 lines
Quoted 6 lines

Perhaps you could test Nick's patches. They are at

http://www.kerneltrap.org/~npiggin/vm/

Against 2.6.2-rc2-mm2. First revert vm-rss-limit-enforcement.patch, then apply those three.

Hmmm, id prefer plain 2.6, but ill try it anyway.

It should go against plain 2.6 with luck.

Applies with offsets against 2.6.2-rc3-mm1-1 and boots. Haven't tested it under high memory pressure yet though.

#5 lore raw
In reply to Samium Gromoff
Quoted 13 lines · Samium Gromoff
Quoted 5 lines
Quoted 2 lines

The machine is a dual P3 450MHz, 512MB, aic7xxx, 2 disk RAID-0 and ReiserFS. It's a few years old and has always run Linux, most recently 2.4.24. I decided to try 2.6.1 and the performance is disappointing.

2.6 has a few performance problems under heavy pageout at present. Nick Piggin has some patches which largely fix it up.

I`m sorry, but this is misguiding. 2.6 does not have a few performance problems under heavy pageout.

It`s more like _systematical_ _performance_ _degradation_ increasing with the pageout rate. The more the box pages out the more 2.6 lags behind 2.4.

Well it is a few problems that cause significant performance regressions. But nevermind semantics...

Quoted 7 lines

What i`m trying to say is that even light paging is affected. And light paging is warranted when you run, say, KDE on 128M ram.

Go measure the X desktop startup time on a 48M/64M boxen--even light paging causes 2.6 to be just sloower. Also the vm thrashing point is much much earlier.

Have a look here: http://www.kerneltrap.org/~npiggin/vm/3/
and here: http://www.kerneltrap.org/~npiggin/vm/4/
patches here: http://www.kerneltrap.org/~npiggin/vm/

and I have a couple of things which improve results even more. True, its only kbuild, but after I do a bit more tuning I'll focus on other things - I'm hoping most of the improvements carry over to other cases though.

Tentatively, it looks like 2.6 under very heavy swapping can actually be significantly improved over 2.4.

Quoted 3 lines

Ask Roger Luethi for details.

Andrew is quite well versed in the details :)