An Ext4 question...

[Dale Farnsworth]

Michael Butash wrote:
> You know, I've heard the same argument against reiserfs for ages, and
> using it on countless servers (both home and enterprise) for at least
> the past 5 years I've _never_ once encountered unrecoverable reiser
> filesystem errors pertaining to whatever kind of ungraceful/ugly reboots
> I've had to do.

Neither have I, though I only used it for a few years.  I evaluated
reiserfs 7 or 8 years ago for MontaVista and helped port it to big
endian systems.  On my recommendation, MontaVista shipped it as their
journalling FS.  Reiserfs was innovative, and in many ways was a major
step forward.

> That and with no manual tweaking of reiserfs, I can say
> I've been pretty darn happy with it.  I've had a great deal more
> occurrences with manually having to fsck ext3 even as rarely as I ever
> do actually use it.

I'm glad and, as I say, my personal experience has been similar.
However, this is unrelated to the issue I mentioned.  There have been
numerous examples of reiserfs users who, after losing some blocks caused
by bad drive hardware, lost vast portions of the reiserfs filesystem.
This results from the B+tree design.  Without backpointers in lower level
FS nodes, the results of the fsck scavenge operation aren't adequate.
IMO, reiserfs isn't as robust as ext3 in the face of hardware errors.

> I was rather quite looking forward to reiser4 before he had to go kill
> his wife...

And I looked forward to it even after.  However, development did slow down.
And, in my opinion, BTRFS development has now surpassed it.  BTRFS hasn't
yet reached the level where I can trust it as much as ext3/ext4, but I have
high hopes for it.  For more info, see <http://lwn.net/Articles/342892/>.

-Dale

> -mb
> 
> On Sun, 2009-08-09 at 11:47 -0700, Dale Farnsworth wrote:
> > > Does anybody know what happens when you stash a huge number of tiny
> > > files in Ext4?  Does it store them efficiently the way ReiserFS does?
> > 
> > No.  Neither ext3 nor ext4 efficiently stores sub-block sized files.
> > The minimum files size granularity is the block size.  The internal
> > fragmentation can hurt with many small files.
> > 
> > As others have mentioned, reiserfs handles that.  However, I can't
> > recommend reiserfs unless extra effort is devoted to backups, since
> > catastrophic failures can result from the loss of a single block
> > in reiserfs.  I have high hopes for btrfs, which appears to solve
> > these issues and more, but I haven't switched over to it yet.
> > 
> > -Dale