You are not logged in.

#1 2009-11-28 09:57:18

flamelab
Member
From: Athens, Hellas (Greece)
Registered: 2007-12-26
Posts: 2,160

Getting ready for kernel26 2.6.32 - Question of ext4 performance

As 2.6.32 approaches, i'd like to post the Phoronix benchmarks for it

http://www.phoronix.com/scan.php?page=a … arks&num=1

There seems to be a "large" performance loss on ext4 due to a specific commit on main kernel branch.

http://www.phoronix.com/scan.php?page=a … ions&num=2

http://git.kernel.org/?p=linux/kernel/g … 25b9aeb745

I'd like to ask if it is possible to have .32 kernel with a patch that would revert that upstream commit.

If that isn't allowed by Arch patching policy, just ignore this thread big_smile

Offline

#2 2009-11-28 10:01:11

wonder
Developer
From: Bucharest, Romania
Registered: 2006-07-05
Posts: 5,941
Website

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

flamelab wrote:

As 2.6.32 approaches, i'd like to post the Phoronix benchmarks for it

http://www.phoronix.com/scan.php?page=a … arks&num=1

There seems to be a "large" performance loss on ext4 due to a specific commit on main kernel branch.

http://www.phoronix.com/scan.php?page=a … ions&num=2

http://git.kernel.org/?p=linux/kernel/g … 25b9aeb745

I'd like to ask if it is possible to have .32 kernel with a patch that would revert that upstream commit.

If that isn't allowed by Arch patching policy, just ignore this thread big_smile

http://wiki.archlinux.org/index.php/Dev … i:Patching

i think kernel developers know better what the hell is going in there smile and should be trusted no?

Last edited by wonder (2009-11-28 10:01:45)


Give what you have. To someone, it may be better than you dare to think.

Offline

#3 2009-11-28 10:10:16

.:B:.
Forum Fellow
Registered: 2006-11-26
Posts: 5,819
Website

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

From the git commit:

We need to flush the write cache unconditionally in ->fsync, otherwise
writes into already allocated blocks can get lost.

Do you really want performance, even at the risk of data loss?


Got Leenucks? :: Arch: Power in simplicity :: Get Counted! Registered Linux User #392717 :: Blog thingy

Offline

#4 2009-11-28 10:14:55

tomk
Forum Fellow
From: Ireland
Registered: 2004-07-21
Posts: 9,839

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

The right way to approach this is to run a comparison yourself using the latest 2.6.32 rc and report the regression on the kernel's bugzilla. That way the right people are informed about it and can decide what to do with it.

Offline

#5 2009-11-28 10:23:27

flamelab
Member
From: Athens, Hellas (Greece)
Registered: 2007-12-26
Posts: 2,160

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

@tomk: Can you provide a reliable benchmark that I could use ? I currently have both .31 and .32-rc8 installed, I could make a test.

Offline

#6 2009-11-28 10:26:18

tverdok
Member
From: Ukraine
Registered: 2009-11-15
Posts: 14

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

You could try IOzone, available in AUR http://aur.archlinux.org/packages.php?ID=22212


"The most likely way for the world to be destroyed, most experts agree, is by accident. That's where we come in; we're computer professionals. We cause accidents."
   Nathaniel Borenstein

Offline

#7 2009-11-28 10:29:16

Ashren
Member
From: Denmark
Registered: 2007-06-13
Posts: 1,229
Website

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

I would say that bonnie++ would fit the bill quite well.

Offline

#8 2009-11-28 10:32:26

flamelab
Member
From: Athens, Hellas (Greece)
Registered: 2007-12-26
Posts: 2,160

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

tverdok wrote:

You could try IOzone, available in AUR http://aur.archlinux.org/packages.php?ID=22212

How can I use it ? man page is full of options.

Trying bonnie++ right now.

Last edited by flamelab (2009-11-28 10:34:16)

Offline

#9 2009-11-28 10:44:24

tverdok
Member
From: Ukraine
Registered: 2009-11-15
Posts: 14

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

flamelab wrote:
tverdok wrote:

You could try IOzone, available in AUR http://aur.archlinux.org/packages.php?ID=22212

How can I use it ? man page is full of options.

Trying bonnie++ right now.

If you decide to use Iozone, just go with automatic mod Iozone -a
I'm not really familiar with bonnie++, so can't help you there.

EDIT: Here is a guide in pdf format http://www.iozone.org/docs/IOzone_msword_98.pdf

Last edited by tverdok (2009-11-28 10:47:44)


"The most likely way for the world to be destroyed, most experts agree, is by accident. That's where we come in; we're computer professionals. We cause accidents."
   Nathaniel Borenstein

Offline

#10 2009-11-28 10:45:51

Ranguvar
Member
Registered: 2008-08-12
Posts: 2,544

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

bonnie++ in main Arch packages is extremely old, I recommend the rock-stable 'experimental' package: http://aur.archlinux.org/packages.php?ID=24114

Offline

#11 2009-11-28 10:47:05

lucke
Member
From: Poland
Registered: 2004-11-30
Posts: 4,018

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

Offline

#12 2009-11-28 10:51:35

tverdok
Member
From: Ukraine
Registered: 2009-11-15
Posts: 14

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

lucke wrote:

Very nice! They even tracked down the cause of the performance drop.


"The most likely way for the world to be destroyed, most experts agree, is by accident. That's where we come in; we're computer professionals. We cause accidents."
   Nathaniel Borenstein

Offline

#13 2009-11-28 10:57:44

flamelab
Member
From: Athens, Hellas (Greece)
Registered: 2007-12-26
Posts: 2,160

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

Bonnie++ from AUR:

System: Intel Core i5 750, 4GB of RAM, WD6401AALS

2.6.31

Version  1.96       ------Sequential Output------ --Sequential Input- --Random-
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
flamepc          8G   723  97 82719   7 39984   4  4192  98 106395   4 385.7   4
Latency             11481us     943ms     265ms    2472us   11436us     306ms
Version  1.96       ------Sequential Create------ --------Random Create--------
flamepc             -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16 24965  33 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
Latency             11731us     422us     437us     112us      82us      81us
1.96,1.96,flamepc,1,1259401928,8G,,723,97,82719,7,39984,4,4192,98,106395,4,385.7,4,16,,,,,24965,33,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,11481us,943ms,265ms,2472us,11436us,306ms,11731us,422us,437us,112us,82us,81us

2.6.32

Version  1.96       ------Sequential Output------ --Sequential Input- --Random-
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
flamepc          8G   822  98 81669   7 38687   3  4198  75 105470   4 255.0   3
Latency             11138us    1600ms    1302ms   74131us   52760us     573ms
Version  1.96       ------Sequential Create------ --------Random Create--------
flamepc             -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
Latency               129us     402us     411us     184us      62us      51us
1.96,1.96,flamepc,1,1259395151,8G,,822,98,81669,7,38687,3,4198,75,105470,4,255.0,3,16,,,,,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,11138us,1600ms,1302ms,74131us,52760us,573ms,129us,402us,411us,184us,62us,51us

I just ran bonnie++ executable, with no options.

I'll try the ozone benchmark now. [edit] iozone -a has a non-readable output hmmm.

Last edited by flamelab (2009-11-28 11:04:42)

Offline

#14 2009-11-28 11:22:34

tverdok
Member
From: Ukraine
Registered: 2009-11-15
Posts: 14

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

Try Iozone -g 4G -Rab [outputfile.wks]


"The most likely way for the world to be destroyed, most experts agree, is by accident. That's where we come in; we're computer professionals. We cause accidents."
   Nathaniel Borenstein

Offline

#15 2009-11-29 21:03:20

R00KIE
Forum Fellow
From: Between a computer and a chair
Registered: 2008-09-14
Posts: 4,734

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

If you know you have no write cache, or that it is safely battery backed, then you can mount with -o nobarrier, and not incur this penalty.

Is this what you want?

But like B says, if they changed it they have a good reason. everyone wants a filesystem that is as fast as possible but data loss is not acceptable at all.


R00KIE
Tm90aGluZyB0byBzZWUgaGVyZSwgbW92ZSBhbG9uZy4K

Offline

#16 2009-12-23 10:16:09

bananaoomarang
Member
Registered: 2009-10-29
Posts: 180

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

well ok we will wait for Btrfs and maybe that will be even more speedy. I want a fast filesystem but data loss would not be very nice and it would NOT be professional of the linux kernel devs to ignore it because tweaking would make the filesystem slightly slower.

By the way flamelab, nice blog.

Offline

#17 2009-12-23 23:13:42

flamelab
Member
From: Athens, Hellas (Greece)
Registered: 2007-12-26
Posts: 2,160

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

I haven't tried yet Btrfs, but from what I see, it has a LOT of potential.

-----
As for the performance regression: it is "eliminated" with barrier=0. But you seriously need to have a UPS then to prevent data loss, especially if you have a lot of data transferred or used or opened to be used.

-------

@bananaoomarang: Thanks tongue

Offline

#18 2009-12-24 00:00:32

akira86
Member
Registered: 2009-01-16
Posts: 119

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

Does ext4 is still faster than ext3 ?

Offline

#19 2009-12-24 16:44:18

R00KIE
Forum Fellow
From: Between a computer and a chair
Registered: 2008-09-14
Posts: 4,734

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

Ext4 became a bit slower in favor of data integrity but the other advantages of ext4 over ext3 still remain, such as faster volume checking if I'm not mistaken (that alone is a big plus for me, checking 100GB ext3 partitions on a notebook can take a good while).


R00KIE
Tm90aGluZyB0byBzZWUgaGVyZSwgbW92ZSBhbG9uZy4K

Offline

#20 2009-12-31 22:25:46

bananaoomarang
Member
Registered: 2009-10-29
Posts: 180

Re: Getting ready for kernel26 2.6.32 - Question of ext4 performance

yeah it's still faster just by a little less than it was before.

Offline

Board footer

Powered by FluxBB