On Friday 07 September 2001 10:41, Guillaume Cottenceau wrote:
> Fabrice FACORAT <[EMAIL PROTECTED]> writes:
>
>
> [...]
>
> > Speed:
> > Despite writing some data more than once, ext3 is often faster (higher
> > throughput) than ext2 because ext3's journaling optimizes hard drive head
>
> The problem with benches is that they are all "tied" to show that "this"
> FS is the fastest. I saw a resembling bench for JFS recently.
>
> So Pixel's bogobench is pretty interesting since it shows more or less a
> typical and no-fs-oriented bench.
>
> And guess what ? the "ext3 is often faster" scales miserably to this
> bench. We can also talk about JFS. Well, they are both winning the prize
> for slowest FS, great ;p.
>
> time df lost
> 3:10 310 + 32 ext2
> 3:35 337 + 1 reiser notail
> 3:50 313 + 4 xfs
> 4:00 300 + 2 reiser (1)
> 4:05 325 + 8 jfs
> 4:05 343 + 31 ext3 (2)
>
>
> ***** DISCLAIMER ****** : I don't care that any FS could be great as
> I-dont-know for server stuff, very large files, sql-throughput, or
> anything. this is a bogobench, nothing more. but personally, considering
> my proper typical use of my machine, I'd choose ext2 (or reiser notail) if
> I consider these results.
Well, the bench I am doing makes 100,000 files, then reads 100,000 times,
then updates 100,000 times, then deletes all the files and times each. The
reads and updates are random and the same file may be chosen many times in
each case. This was designed to simulate mailserver activity which is
typically one of the most stressful uses of filesystems. (And my
benchmarking activity has as secondary purpose breaking the filesystem if it
is breakable).
My results thus far, with create/delete, follow Pixel's very closely. I
should have full info late saturday (takes some time to do this).
JFS is _still_ not problem free. As a normal partition, it seems mostly
stable, but as a root partition, it is a formula to wreck a system. At least
it passes the sledgehammer on non-root partitions.
Civileme
