How dangerous is it to run xfs without write barriers?

On 12/22/08, Chris Robertson <crobertson AT gci DOT net> wrote:
> dan wrote:
>> I guess that updatedb thing reinforces my arguement about not seeing
>> any mixed load tests.  ext3 handles these situations pretty good,
>> maybe XFS does not...
>
> Write barriers really harmed XFS performance on my setup (16 Seagate
> ES.2 spindles attached to an Adaptec 51645 utilizing hardware RAID6).
> iostat was showing a peak of 400 tps with barriers.  Mounting nobarrrier
> raised that limit to to over 20,000.  Obviously your mileage may vary.
> Two interesting data points to note, it appears that LVM doesn't support
> barriers
> (http://hightechsorcery.com/2008/06/linux-write-barriers-write-caching-lvm-and-filesystems),
> and ext3 (and ext4) don't use barriers by default
> (http://lwn.net/Articles/282958/).  The design allows for this without
> as much risk as might be expected (http://lwn.net/Articles/283168/).
>
> Back to XFS, allocation groups, unless specified at file system creation
> are calculated on file system size, and can have a great effect on
> performance when multiple threads are contenting for FS access.
> Changing the journal size can also have an effect on performance, but
> again, this is only possible at creation.  Changing the number  and size
> of log buffers is a mount time modification, and might also have a
> decent effect on performance.  The kernel documentation has more
> information on this.
>
>>
>> By the way, I read that EXT4 should allow for EXT3>EXT4 upgrades.
>
> Same thing for btrfs.  Neat stuff.
>
>>   One(of many) nice things about EXT4 is delayed writes which
>> essentially means write re-ordering to mask/reduce I/O bottlenecks.
>
> Which XFS already has...  But they are affected by write barriers.
>
>>   Hopefully EXT4 will become stable pretty soon!
>
> Agreed.  As a side note, development on Reiser4 is ongoing:
> http://marc.info/?l=reiserfs-devel&r=1&w=2
>
> Finally, if you are sleeping too well at night because you think your
> data is safe, I have a couple of papers your might be interested in that
> I stumbled across while fact checking:
>
> Model-Based Failure Analysis of Journaling File Systems (covers ext3,
> Reiserfs, and JFS):
> http://www.cs.wisc.edu/adsl/Publications/sfa-dsn05.pdf
>
> Failure Analysis of SGI XFS File System:
> http://pages.cs.wisc.edu/~vshree/xfs.pdf
>
> Chris
>
> ------------------------------------------------------------------------------
> _______________________________________________
> BackupPC-users mailing list
> BackupPC-users AT lists.sourceforge DOT net
> List:    https://lists.sourceforge.net/lists/listinfo/backuppc-users
> Wiki:    http://backuppc.wiki.sourceforge.net
> Project: http://backuppc.sourceforge.net/
>


-- 
http://resc.smugmug.com/

------------------------------------------------------------------------------
_______________________________________________
BackupPC-users mailing list
BackupPC-users AT lists.sourceforge DOT net
List:    https://lists.sourceforge.net/lists/listinfo/backuppc-users
Wiki:    http://backuppc.wiki.sourceforge.net
Project: http://backuppc.sourceforge.net/