Subject: Re: write cache on ATA drives
To: Tld <tld@tld.digitalcurse.com>
From: Sean Davis <dive@endersgame.net>
List: tech-kern
Date: 12/05/2002 13:32:48
On Thu, Dec 05, 2002 at 06:51:53PM +0100, Tld wrote:
> Sean Davis wrote:
> >the cache issue, and have never had any problems because of it. Should I
> >disable write caching, even though I've never had it cause a problem?
>
> I'd rather disable soft dependencies.
It's already disabled on every partition I can disable it on without
dropping to singleuser. Also, I've never used softdeps on a partition that
contained critical data - so never on important things like /, /var, /usr,
etc..
> Do some benchmarking, with and without softdeps. If the drive is reordering
> writes you should see almost no difference.
Hmm, on the WD drive (the special edition one I have has 8MB of cache on it)
it seems to perform better without softdeps:
(with softdeps)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
100 18650 91.7 24160 59.0 34755 75.6 19176 96.6 103726 97.0 5470.9 97.3
(without)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
100 21915 91.6 35532 53.1 35816 76.8 18877 95.6 104340 95.7 5360.9 91.9
only a small drop in sequential input and random seeks for no softdep, which
isn't a big enough difference to concern me.
> A big difference should happen in case you disable write cache on the disk.
>
> And, last time I checked, softdeps were still unstable in certain
> conditions.
which is why I never use them on a partition I really care about ;)
-Sean
--
/~\ The ASCII
\ / Ribbon Campaign Sean Davis
X Against HTML aka dive
/ \ Email!