On Mon, May 24, 2004 at 03:03:36PM +1000, Pietro Abate wrote:
> Hi,
>
> On Mon, May 24, 2004 at 12:46:53PM +0800, Yusuf Goolamabbas wrote:
> > Are the above statements only true for software raid5 or true for all
> > software raid levels. Are there any other values appropiate for software
> > raid/hardware raid ?
>
> in my understanding the big problem with raid5 is that most controller
> have a fix block size, but xfs uses two different bs for logs and data.
Theres a bit more to it than that, but thats close to the mark, yes.
> Having external logs can alleviate the problem.
External logs help, but don't completely alleviate the problem.
Using a sector size that matches the filesystem blocksize is the
key, as it ensures _all_ IOs coming out of XFS are aligned with
that size boundary.
> However it's still not clear to me how to tweak performance of xfs both
> on raid1/5 and avoid these pitfalls. Moreover the xfs website says that
> xfs performance on raid are poor and the problem is going to be solved
> in the next major release.
The website is quite out of date & needs lots of updating, so
I wouldn't put too much faith in what it says at the moment.
Unfortunately we're always too busy hacking the code to update
it.
cheers.
--
Nathan
|