Re: Will we ever see XFS supported in Red Hat Enterprise Linux?

[Jon Lewis <jlewis@xxxxxxxxx>]

On Mon, 29 Aug 2005, Bryan J. Smith wrote:

> Actually, that would be nice since Red Hat Linux 7.3 is still
> supported by FedoraLegacy.ORG, Progeny and select, mission
> critical hardware/software companies like FalconStor.
>
> Then again, the Red Hat Linux 9 / XFS 1.3.1 releases _are_ on
> the SGI site.  I only had 1 system running with those, and
> have far less time with them, but they were fairly good
> AFAICT.

AFAIK, putting a Red Hat 9 kernel on an 8.0 system is going to cause 
trouble (NPTL will cause problems), but I suspect I'll have less trouble 
using a 7.3 kernel on an 8.0 system.  I could be wrong.  I should probably 
download SGI's RH9 srpm and see if it can be made to work without too much 
trouble on a 8.0 based system.

> As I mentioned off-list, Red Hat Linux 8.0 is a ".0" release
> and not well trusted.  And your issues are compounded by the
> fact that you are running a 3rd party "hacked together" XFS
> release.

The servers we have running 8.0 are highly specialized and are running 3rd 
party software for those functions anyway (qmail based mail servers), and 
so they really don't rely on much of the distro other than the kernel. 
I'm not too concerned with distro version...just kernel/fs stability. 
That's why I tried the 2.4.31 kernel from SGI's xfs-cvs, but I found its 
nfsd not entirely functional.

If I'm going to have to shut down long enough for an OS upgrade, and then 
deal with anything broken by the upgrade, I'd rather just swap out the 
disk array for another running ext3, and then copy all the old mail in, 
with much less "completely down" time.

> I haven't been installing Opteron systems with less than 4GiB
> of RAM since they first came out.  Heck, 16GiB RAM is pretty
> much standard when I install a 4-way Opteron.

But how much RAM is required to xfs_repair a 5TB fs?  Does anyone even 
know?  How about a 25TB fs?  Even if XFS allows you to create such fs's, 
it doesn't mean you're not much better off doing it the old fashioned way 
and tying several smaller fs's together via directory structure rather 
than creating one giant one.  What sort of data are your clients storing 
that they "need" such large fs's?

> I assume you're referring to the previous integrator, not
> SGI, XFS or anything else.  I tire of fly-by-night system
> integrators who put in hacked kernels and ".0" releases.

Not an integrator...ex-staff/coworkers choosing to use unsupported FS's 
requiring 3rd party "hacked together" kernels.

----------------------------------------------------------------------
 Jon Lewis                   |  I route
 Senior Network Engineer     |  therefore you are
 Atlantic Net                | 
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________