On Mon, Jan 03, 2005 at 04:03:25PM +0100, Johnny Billquist wrote:
> On Mon, 3 Jan 2005, Martin Husemann wrote:
>
>> How meaningfull is such a benchmark on different hardware?
>
> Not to mention the unknown possible other loads on the machine. It's not 
> even measuring how much cpu-time have been spent, but it's just a 
> measurement of the actual time for the command to complete.

Not only that, it doesn't take into account the state of the filesystem
when you started the command.  Did you start with a freshly newfs'ed
filesystem each time?  What about block-sizes, etc?  An older, actively
used filesytem will be more prone to fragmentation (stuff scattered all
over the disk/filesytem) than one less utilized and/or just had newfs
run on it.

About 6 months ago, I moved my homedirs on this box to a new hard-drive.
After the newfs, I used a dump/restore pipeline to copy over the files.

Before the copy, it took nearly 20 seconds to open my
+lists/netbsd/users nmh folder in mutt with over 1000 msgs, afterwards,
just a couple of seconds.  Now, its slowed down again, lots of activity,
lots of little files getting created and deleted, over time, things
scatter, get less efficient.

This would be the similar under FFS, ext3, xfs, lfs, reiser, whateverfs.

-- 
Michael Parson
mparson@bl.org