Federico Sevilla III wrote:
>
> On Mon, 23 Jul 2001 at 16:01, [EMAIL PROTECTED] wrote:
> > I am looking for some documentation relating to the differences
> > between the main 3 journaling filesystems, and why some would be more
> > appropriate in some situations than others.
>
> Comparing the three will be a rather difficult task. I know because I've
> tried. And while I'm happy with the results of my personal research, I'm
> sure I'm missing out on a variety of points and do not have the "universal
> truth".
>
> You may start things out by checking the benchmarks that Constantin
> Loizides recommended. The second, in particular, seems to be pretty okay.
We have not yet reproduced those benchmarks, and I find them inconsistent with our
benchmarks.
There is not enough information on the website to reproduce the benchmark. Hopefully
Elena will
produce some result soon though.
I suspect that they did not use -notails. ReiserFS can give you much better space
efficiency (using
tail combining), or it can give you better performance (with -notails specified, and
gaining you
space savings in inode allocation but not from tail combining), but it can't give you
both better
performance and space savings from tail combining (not until v4.... ).
> There are also benchmarks listed in the ReiserFS website, but these don't
> seem to be universally accepted because of the average size of the files.
> (Too small for what most others would consider normal).
Is 100k too small? We have three different sections of the benchmark, one small
files, one medium
files, one large files. Please review the whole benchmark in more detail, there is
more to it that
you might guess.
We follow the 80/20 rule as embodied in reiser_fract_tree.c, which all real
filesystems follow and
no other benchmarks follow.
Hans
