Hi,

Kouichirou Hiratsuka --> current-users (2006-04-09 18:27:42 +0900):
> Hello,
> 
> On my machine (ThinkPad X32), disk write performance with softdep
> is about 4 times as slow as before.  It takes about 3 or 4 times
> as long to copy the files.  And with bonnie++, the sequential
> output (per char) is 1/4.
> 
> It seems to be related to this commit (please see the following
> results of bonnie++).
> 
>   http://mail-index.NetBSD.org/source-changes/2006/03/30/0021.html
[...]
> Anyone else seeing this?

Yes, I am. I just ran bonnie++ on a ffs (on a ccd(4) striping two
partitions on two IDE disks, both of them having write-back cache
enabled) mounted with softdeps. The 'Sequential Output' 'Per Chr'
results were indeed worse than expected:

	- 4733 K/sec with a recent GENERIC kernel
	- 66285 K/sec with sources from just before the commit you quote

(average of three runs each). That's worse by a factor of 14!

It's only the `Putc' test which performs badly. All other tests return
about the same results with both kernels, and on a ffs mounted without
softdeps even the `Putc' test results are on a par.


Regards, Jukka

-- 
bashian roulette:
$ ((RANDOM%6)) || rm -rf ~