lkml.org 
[lkml]   [2008]   [Jun]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Performance of ext4
On Wed, Jun 11, 2008 at 08:02:32AM +0000, Holger Kiehl wrote:
> Hello
>
> Doing some performance test between ext3 and ext4 I noticed that ext4
> is not much faster or in some cases slower then ext3. Two years ago when
> I tested ext4 it was a lot faster then ext3 (see my mail:
> http://lkml.org/lkml/2006/6/6/65). Doing some simple tests with bonnie++
> I got the following results:
>
> Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
> -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
> ext3(2 years ago)16G 38621 98 194816 94 87776 49 37921 92 239128 54 1402 5
> 16G 47000 99 194276 94 89232 49 38628 92 240539 55 1399 5
> 16G 45873 98 178195 90 89726 50 38482 92 240490 55 1381 4
>
> ext3 (now) 16G 51501 97 210601 91 100479 32 55528 98 301589 44 1198 5
> 16G 52702 98 215540 94 99339 32 55376 97 300933 44 1159 4
> 16G 52426 99 212584 94 99091 31 55656 98 301669 44 1160 4
>
> ext4(2 years ago)16G 59223 91 264155 45 111459 36 57313 99 317944 63 1478 7
> 16G 58814 92 276803 47 110418 36 57105 99 317534 65 1525 5
> 16G 58299 92 274523 48 110290 36 56723 99 318839 65 1502 4
>
> ext4 (now) 16G 52965 98 224199 89 108440 32 56389 99 303792 42 1526 4
> 16G 52792 98 223980 92 107685 32 56350 98 303066 42 1532 4
> 16G 52994 98 222354 92 107802 32 56386 99 303727 41 1455 4
>
> For this system the write performance is the most important factor and one
> can see today ext4 is marginally faster then ext3. But 2 years ago ext4 was
> a lot faster (~270MB against ~223MB).
>
> Using my own benchmark afdbench where many process copy thousands of small
> files around the results are as follows:
>
> For ext3: 5449.76 files per second 15.58 MiB/s
> For ext4: 5162.16 files per second 15.48 MiB/s
>
> So in this test ext4 is a bit slower then ext3. Since afdbench has seen
> considerable changes, one cannot compare these results with those 2 years
> ago. But 2 years ago ext4 was 12% faster then ext3.
>
> Test where done with kernel 2.6.25.4 and file system where created as follows:
>
> ext3: mke2fs -b 4096 -m 0 -O dir_index,large_file,filetype,has_journal,sparse_super -j /dev/md7
> ext4: mke2fs -b 4096 -E test_fs -m 0 -O dir_index,large_file,filetype,has_journal,sparse_super -j /dev/md7
>
> And both where mounted with the following options:
>
> noatime,nodiratime,commit=15
>
> 2 years ago I used 2.6.16.8 but the hardware is still the same. So what has
> happened with the performance of ext4? I noticed that 2 years ago I could
> use extents+mballoc+delalloc, now there is only extents+mballoc in the
> current kernels. Could delalloc make the big difference? I saw that
> in Andrew Morton mm tree delalloc is included. Unfortunately when I tried using
> 2.6.26-rc2-mm1 a sync would never return and there where lot of other
> odd things, so I could not do any tests with delalloc.


The sync and other related hangs should be fix with the latest patches
vailable at http://repo.or.cz/w/ext4-patch-queue.git. Using mballoc have
an impact on CPU utilization because we try to build an in-memory extent
map of free blocks available in the group. The cold cache run (the first
run) would take more time because of the time needed to build the extent
map. So repeating the same test and looking at the numbers would help us
understand the impact of in-memory extent building code.

>
> So any idea what I am doing wrong or what I could do to improve those numbers?
> Please CC me since I am not subscribed to the list.
>

You should be able to apply the patches in the patchqueue mentioned above to 2.6.26-rc5
Can you test with the same and get the numbers. ? Also delalloc enabled
by default with changes from patchqueue force writeback mode for
journal. So you may want to enable writeback for ext3.

-aneesh


\
 
 \ /
  Last update: 2008-06-11 13:07    [W:0.112 / U:0.176 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site