Because XFS is BETTER (Re: opinion on XFS)

Joshua Baker-LePain jlb17 at
Thu May 9 07:08:11 PDT 2002

On Thu, 9 May 2002 at 9:24am, Mark Hahn wrote

> > The conventional wisdom is that XFS performs best of the Linux FSs for 
> > streaming large files to/from disk.  After all, that's what it was 
> > designed for (on IRIX).
> hmm, but ext2 has always been able to deliver platter-leven bandwidth,
> as well.  could it be that people got a bad impression of ext2 by
> configuring it with 1k blocks?

I should have said "handling large files" rather than just the particular 
case of streaming them to/from disk.

> >
> > (Translated from the original spanish -- sorry for the long link).
> I think "transmunged" might be more appropriate ;)

Yep -- that's always the case.  But it's a better translation than I could 
do.  :)

> it was interesting, but a little unclear on methodology.

Yeah, it is.  Of course, I can always find/come across these things when 
I'm not looking for them.  But when you really want them...

> > Disclaimer:  This all comes from someone with a 560GB (and 93% full -- 
> > damn pesky grad students) XFS formatted RAID exported to his cluster who 
> > has been *very* happy with the performance and stability of said 
> > filesystem.
> that's great.  but is there any reason to think ext3 wouldn't behave as well?

Not really.  But when I was testing this thing and putting it into 
production last March/April, ext3 for 2.4 wasn't there yet.  The only two 
options were Reiser and XFS.  XFS got the nod (even though 1.0 wasn't 
out yet) since our data sets are large and it didn't have a history of 
issues with NFS.  The option of fully functional ACLs is a nice one as 

And with 518GB on disk now, reformatting isn't exactly an option.  :)

Joshua Baker-LePain
Department of Biomedical Engineering
Duke University

More information about the Beowulf mailing list