2014-12-12 06:03:54

by alex chen

[permalink] [raw]
Subject: A doubt on journal_async_commit option

This commit 0e3d2a6313(ext4: Fix async commit mode to be safe by using
a barrier) show that using journal_async_commit feature has a 50%
performance improvement. But I tested in SUSE Linux Enterprise Server
11 SP3(linux kernel 3.0.93) and Red Hat Enterprise linux 6.4(linux
kernel 2.6.32), the result show this feature has no performance
improvement.
My test command:
mount /dev/sdb /mnt/sdb
./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
umount

mount /dev/sdb /mnt/sdb -o journal_async_commit
./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
umount

My test result:
FSUse% Count Size Files/sec App Overhead
6 1000 10240 42.1 10671
vs.
-o journal_async_commit
FSUse% Count Size Files/sec App Overhead
6 1000 10240 63.9 10625

Superblock info(-o journal_async_commit):
Filesystem features: has_journal ext_attr resize_inode dir_index
filetype needs_recovery extent flex_bg sparse_super large_file
huge_file
uninit_bg dir_nlink extra_isize
Block size: 4096
Journal backup: inode blocks
Journal features: journal_checksum journal_async_commit
Journal size: 128M
Journal length: 32768
Journal sequence: 0x000017a6
Journal start: 1

Am I missing something?

---Alex



2014-12-12 14:54:51

by Theodore Ts'o

[permalink] [raw]
Subject: Re: A doubt on journal_async_commit option

On Fri, Dec 12, 2014 at 02:03:32PM +0800, alex chen wrote:
> This commit 0e3d2a6313(ext4: Fix async commit mode to be safe by using
> a barrier) show that using journal_async_commit feature has a 50%
> performance improvement. But I tested in SUSE Linux Enterprise Server
> 11 SP3(linux kernel 3.0.93) and Red Hat Enterprise linux 6.4(linux
> kernel 2.6.32), the result show this feature has no performance
> improvement.
> My test command:
> mount /dev/sdb /mnt/sdb
> ./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
> umount
>
> mount /dev/sdb /mnt/sdb -o journal_async_commit
> ./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
> umount
>
> My test result:
> FSUse% Count Size Files/sec App Overhead
> 6 1000 10240 42.1 10671
> vs.
> -o journal_async_commit
> FSUse% Count Size Files/sec App Overhead
> 6 1000 10240 63.9 10625

Um, the files per second went up from 42.1 to 63.9 --- that's a 50%
improvement, yes?

- Ted

2014-12-15 03:48:07

by alex chen

[permalink] [raw]
Subject: Re: A doubt on journal_async_commit option

On 2014/12/12 22:54, Theodore Ts'o wrote:
> On Fri, Dec 12, 2014 at 02:03:32PM +0800, alex chen wrote:
>> This commit 0e3d2a6313(ext4: Fix async commit mode to be safe by using
>> a barrier) show that using journal_async_commit feature has a 50%
>> performance improvement. But I tested in SUSE Linux Enterprise Server
>> 11 SP3(linux kernel 3.0.93) and Red Hat Enterprise linux 6.4(linux
>> kernel 2.6.32), the result show this feature has no performance
>> improvement.
>> My test command:
>> mount /dev/sdb /mnt/sdb
>> ./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
>> umount
>>
>> mount /dev/sdb /mnt/sdb -o journal_async_commit
>> ./fs_mark -d /mnt/sdb/ -s 10240 -n 1000
>> umount
>>
>> My test result:
>> FSUse% Count Size Files/sec App Overhead
>> 6 1000 10240 42.1 10671
>> vs.
>> -o journal_async_commit
>> FSUse% Count Size Files/sec App Overhead
>> 6 1000 10240 63.9 10625
>
> Um, the files per second went up from 42.1 to 63.9 --- that's a 50%
> improvement, yes?
>
> - Ted
Yes, I understand it. Thank you!

> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>