Re: So: reiserfs or ext3 (was Re: ext3 to be in 2.4.15!)

[Eduard Bloch <edi@gmx.de>]

#include <hallo.h>
Karsten M. Self wrote on Sat Nov 10, 2001 um 02:48:33PM:
> > Now that you said this... I'd like to see how reiserfs and
> > ext3 in writeback mode perform. I'd think ext3 would still be
> > outperformed by reiser for the large dir listings.
> 
> The issue isn't writing the directories, but searching them.  Hash beats
> linear scan, above the hash overhead threshhold.  This crossover's
> clearly evident for 10,000 directory entries, and is probably
> significant at substantially fewer.

Please first how the thing works before adjucating relying on the traditional
prejuducises (like speed: hash  >> tree >= list). This ist not allways true if
you know how the client program work. In case of ext2, there were allways a
problem: when doing file access in a directory (readdir), the ext2 driver did
allways start at the top of the list and searched for that entry. Ext3 stores
the position the last found entry, so when accessing again, the search begins
an this position and you get your entry wery fast, in best case within the
first search loop.
Result: many traditional programs which read the directory contants in their
"natural" order (and sort them later on displaying) work dimensions faster in
large full directories than with the default ext2 driver. Btw, this
modification could also be ported to ext2, I wonder why did the people not try
to optimise ext2 earlier instead of developing ReiserFS. 

Gruss/Regards,
Eduard.