[wellylug] Convert ReiserFS to Ext2

[David Zanetti]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Mon, 5 Apr 2004, Glen Ogilvie wrote:

> On the subject of file systems.  Does anyone have anything bad to say about 
> XFS?  I am considering using it.

Prior to integration, it apparently wasn't too bad, and there are some
example sites which are pretty large (eg, Weta Digital) who were using
it. But I suspect they were getting some pretty good support for it :)

Post-2.6 intergration, it seems like a crapload of problems have been
uncovered, possibly due to the quite different lower layers in 2.6 vs
2.4. Some systems I'm involved in which seem to get corruption a lot more
often for an alledgely "stable" filesystem. Certainly more than I've
experienced with ext3, in any case.

One thing which is missing from this rather opinionated debate is the
difference between stablity in terms of how well the code is known, and
stablity in adverse conditions. They are not the same thing. You may have
many situations where in adverse conditions (eg, actual disk failure, or
the box being kicked over, etc) where a filesystem design performs better
than other designs. Journalling filesystems will generally do better than
non-journaling, just because the metadata is in a known state. 

But that's hugely different to trusting that the _implementation_ is
correct. Ext2 has huge problems in adverse conditions, but it's extremely
well understood and validated code. It's stable, in the sense that the
code is known to be good against the design. Look at the change logs, you
don't see huge patches going in against stable kernels to fix ext2
corruption. 

Lastly, there is _no_ one size fits all. Do some research about the
workloads, and the design choices made for each filesystem. Think about
what you're actually using the filesystem for. Filesystems designed for
lots of small files make little sense for, say, my PVR which almost always
only ever creates files of roughtly 500M clumps. Using a journaling
filesystem for a real RDBMS (eg, Postgres) is just wasting resources since
metadata is maintained by the database anyway, and there are even cases
where the extra journaling provided by the filesystem can lead to data
corruption. Is being able to mount it on older systems or rescue
situations important? These are important questions to have some idea on,
before blindly asserting randomFS is "best".

IMO.

- -- 
David Zanetti           |  (__)             
#include <geek/unix.h>  |  ( oo    Mooooooo 
http://hairy.geek.nz/   |  /(_O ./         
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: Made with pgp4pine 1.75-6

iD8DBQFAcTrVT21+qRy4P+QRAht3AJ9bA/bdtT6Nd/ZPLEuygHtJ+xJ5oQCfTBY1
798Zg5syax4urCpGwLZ9zEI=
=6G1f
-----END PGP SIGNATURE-----