[Top] [All Lists]

Re: TAKE - write performance improvement

To: Jens Axboe <axboe@xxxxxxx>
Subject: Re: TAKE - write performance improvement
From: Russell Cattelan <cattelan@xxxxxxxxxxx>
Date: Tue, 29 Aug 2000 18:35:40 -0500
Cc: Steve Lord <lord@xxxxxxx>, Ananth Ananthanarayanan <ananth@xxxxxxxxxxxxxxxxx>, linux-xfs@xxxxxxxxxxx
References: <lord@sgi.com> <200008291644.LAA04616@jen.americas.sgi.com> <20000830011048.B6674@suse.de>
Sender: owner-linux-xfs@xxxxxxxxxxx
Jens Axboe wrote:

> On Tue, Aug 29 2000, Steve Lord wrote:
> > Looks like the corruption is once again coming out of the clustering write
> > code in pagebuf, I would recommend not using the kiocluster mount option
> > at the moment. I am doing a glibc build with Ananth's change back in again
> > and put it back into the tree if the build works OK.
> Aughr, this could explain some weirdness that I'm seeing on IDE kiobuf
> atm. Basically, kio and "plain" xfs works just fine and I can't crash
> it. No corruption either. But kiocluster corrupts data here too...
> And currently I don't see why. For IDE, the kio clustering just means
> that the same request will be started more than one time. I'll try and look
> closer at this tomorrow.

It is unclear at this point where the corruption is coming from
there is evidence that it has nothing to do with kiobuf's/ clustering.

LVM testing is also showing corruption... it doesn't use kio bufs.


<Prev in Thread] Current Thread [Next in Thread>