[Top] [All Lists]

Re: defrag xfs

To: Chris Wedgwood <cw@xxxxxxxx>
Subject: Re: defrag xfs
From: Sonny Rao <sonny@xxxxxxxxxxx>
Date: Fri, 21 Jan 2005 02:08:17 -0500
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <20050121064532.GA28594@xxxxxxxxxxxxxxxxxxxxx>
References: <F62740B0EFCFC74AA6DCF52CD746242D010337FA@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx> <41F07494.1060501@xxxxxxx> <20050121043237.GA28699@xxxxxxxxxxxxxxxxxx> <20050121051228.GA28161@xxxxxxxxxxxxxxxxxxxxx> <20050121055051.GB29637@xxxxxxxxxxxxxxxxxx> <20050121064532.GA28594@xxxxxxxxxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/
On Thu, Jan 20, 2005 at 10:45:32PM -0800, Chris Wedgwood wrote:
> On Fri, Jan 21, 2005 at 12:50:51AM -0500, Sonny Rao wrote:
> > Yeah, my apologies if this question has been answered before, but as
> > you point out, this is something that is desired by a great many
> > people.
> >From what I can tell, it comes up often from people playing with XFS
> on their desktop (because it's cool and has more buzzwords than ext3).
> People generally resize to bigger sizes, because they accumulate data
> for various reasons, grow the business or whatever.
> I really don't hear many good reasons for shrinking the fs other than
> "I didn't think about the sizes I needed and now /pr0n could really
> use some of the space I'm not using in /var".
> I suspect if there was any real commerical interest in this
> (ie. someone would fund it's development) there are about a dozen
> people who could take a crack at this right now.

Sure, I agree with everything you say here.  It just bugs me from a
completeness standpoint that XFS and JFS can't do it right now.  Call
it an itch.

> > Far be it for me to pertend I understand all of the issues involved,
> > but as a project I'd find it an interesting one to at least work on.
> Steve Lord has given a description of what is required somewhere on
> the list.  I also have something I made somewhere on how this might be
> done without too many kernel changes.
> > Maybe just a "simpler" off-line version would be easier to write,
> > and wouldn't require kernel changes?
> My initial thoughts are that would probably be quite a bit harder as
> you would need to use libxfs to do all sorts of fs-magic in userspace
> that we already do in the kernel with pretty well tested code-paths.

Cool, I'll search the archives, thanks.


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