Re: Performance of ext4

[Holger Kiehl <Holger.Kiehl@dwd.de>]

On Tue, 17 Jun 2008, Holger Kiehl wrote:

> Doing several test with '-m 0' I was unable to reproduce this and I could
> now do several runs with afdbench. However the results do show that with
> ext4-patch-queue it actually slower:
>
> For ext3:             5449.76 files per second 15.58 MiB/s
> For ext4:             5162.16 files per second 15.48 MiB/s
> For ext4+patch-queue: 4963.6975 files per second 14.73 MiB/s
>
> On the positive side the bonnie++ numbers got much better:
>
> 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            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            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
> ext4(patchqueue)16G 59727  98 252733 52 110177 25 55821  98 296739 42  1553 
> 5
>                16G 61047  99 239242 48 111664 25 55706  98 297151 42  1545 
> 4
>                16G 60503  99 241532 47 109655 25 55671  98 297648 42  1552 
> 3
>
> ext3 and ext4 tests where done with 2.6.25.4 and those with patch-queue was
> 2.6.26-rc5. I will do another test run with 2.6.26-rc5 without patch-queue
> just to make sure that the slowdown does not happen due to changes in the
> 2.6.26 branch.
>
Here the results without patch queue:

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
                 16G 52133  98 221378 95 106873 32 55707  99 297065 42  1546   4
                 16G 52042  98 220931 93 107715 32 55939  98 298810 42  1543   3
                 16G 52975  98 220976 93 108060 31 56426  98 298906 42  1452   4

For afdbench: 5336.41 files per second 15.63 MiB/s

So it seems that for afdbench the ext4-patch-queue is a slowdown.

I forgot to mention that for bonnie ext4-patch-queue reduces CPU-load
a lot. For block writting it is nearly halved.

Holger