mongo was designed to show me what file size ranges reiserfs performance was
weak at. Yes, if you
look at one part of it, it shows you where reiserfs is fast at small files, but
if you look at
another part of it, it shows you where reiserfs is slower than XFS for very
large files. We'll try
to fix that in reiser4.....:-) (I think I know exactly what the problem is, we
may even fix it in
v3.)
I think the most significant feature of mongo.pl though, is that it attempts to
generate a realistic
(and tunable) fileset that reflects the old 80/20 rule that 80% of the bytes
are in 20% of the
files, and it uses a fractal file size distribution function that I think is
innovative in
benchmarking. We should modify mongo.pl to both benchmark using that rule, and
to benchmark with
rigid file size ranges. Elena, I'd like to see that.
XFS is a great filesystem, and reiser4 learns much from it.
Hans
|