Matt,
Thanks for the reply. I have already made my RAID5 software arrays and they
have data on them so I can't remake the filesystem now. But can you tell me if
there's a way to check what the sector size is on the current filesystem? I
ran mkfs.xfs with the block size = 4096. I didn't specifiy a sector size. Is
the default also 4096?
Andy Liebman
In a message dated 1/5/2004 2:11:09 PM Eastern Standard Time, matts@xxxxxxx
writes:
I don't know about anything specific to 2.6, but I know that software
RAID5 doesn't work as smoothly with XFS as with ext2/3 or reiserfs. Since
XFS uses sector size for log writes and block size for data writes, make
sure they're the same size on Linux software RAID5. Otherwise the RAID
system will be constantly flushing it's cache, which will slow things down
a lot. So if your filesystem will have 4k blocks, make it with:
mkfs.xfs -d size=4096 -s size=4096 ... /dev/md0
This was under the 2.4 kernel, but I don't think that part of the RAID
system has changed with 2.6.
--
Matt Stegman
On Sat, 3 Jan 2004 AndyLiebman@xxxxxxx wrote:
> Is there any "definitive word" about whether it's safe to use xfs under the
> released version of the 2.6 kernel? Also, is there anything to watch out for
> using xfs plus software RAID5? If anybody has this information, it might be
> nice to post it.
>
> Andy Liebman
|