<4> post_create: setxatter failed

[Dan Hollis]

On Mon, 23 May 2005, Arjan van de Ven wrote:
> On Sun, 2005-05-22 at 23:50 -0700, Dan Hollis wrote:
> > On Mon, 23 May 2005, Arjan van de Ven wrote:
> > > "linux jfs" isn't documented afaik.
> > > The simple situation is that ext3 is basically all we really support and
> > > test, the rest may or may not work. Is there a reason you want to use JFS ?
> > > (ext3 in fc3/fc4 is pretty competative with any of the other filesystems on
> > > just about every workload performance wise... the benchmarks I've seen from
> > > others hardly ever put JFS on top for anything nowadays so JFS strikes me as
> > > a bit of an odd choice)
> > xfs and reiserfs are _huge_ wins over ext3 for news servers.
> 1) Did you try this on 2.4 or 2.6? 2.6 ext3 (with htree and
> reservations) is like a 3x improvement over the 2.4 ext3 in many
> workloads and is sometimes even slightly better than reiserfs in the
> "milions of files in a directory" scenario.

Both.

Its not just for large directories, reiserfs did much better with many 
small files too (typical of news and mailservers). Note that reiserfs 
wins _big_ in this case performance wise if you turn off tailmerging. In 
my tests reiserfs lost vs ext3 on many-small-files-writing if you had 
tailmerging on, but this was a tradeoff for ~10% or more extra storage 
space you got from tailmerging.

> 2) Did you set the ext3 journalling mode to be on par with reiserfs/xfs?
> (By default ext3 uses a more strict journalling mode to increase data
> integrity but that costs some performance vs reiserfs and xfs that don't
> have this extra protection)

I tried all available ext3 journalling modes, it was never faster and 
often many times slower.

But even with non-strict journaling we got some total filesystem 
failures with ext3 where we did not with reiserfs and xfs on identical 
systems with identical hardware and identical workloads.

We still have a few ext3 machines about but they are being phased out and 
replaced with reiserfs as we get the chance.

-Dan