Subject: Re: NEW_BUFQ really cool!
To: None <current-users@netbsd.org>
From: Takahiro Kambe <taca@back-street.net>
List: current-users
Date: 04/01/2003 00:31:45
In message <20030331084303.GF746@phaeton.entropie.net>
on Mon, 31 Mar 2003 10:43:03 +0200,
Martin Weber <Ephaeton@gmx.net> wrote:
> > I've lost my local CVS repogitry once. You should check it under
> > heavy file system access.
>
> I'm using it successfully for some time, and also under high load. You
> said later it was in conjunction with softdeps, well, I don't use softdeps,
> for I don't consider the softdeps stable. I consider NEW_BUFQ on its own
Hmm, it is too bad that a feature containted in release version
(i.e. NetBSD 1.6) isn't reliable than containted in current, unless it
is described in somewhere.
At least, sys/arch/i386/GENERIC's comment looks SOFTDEP is safer than
NEW_BUFQ_STRATEGY.
options SOFTDEP # FFS soft updates support.
# Enable experimental buffer queue strategy for better responsiveness under
# high disk I/O load. Use it with caution - it's not proven to be stable yet.
#options NEW_BUFQ_STRATEGY
> as stable though. And you don't need to enable NEW_BUFQ on a softdeps
> partition to get crashes and data loss :)
With my understanding, SOFTDEP is file system level feature and
NEW_BUFQ is device driver level. I think that NEW_BUFQ_STRATEGY isn't
enable/disable per partition.
Anyway, using both might result a bad influecnce in performance, it
shouldn't crash. ;-(
--
Takahiro Kambe <taca@back-street.net>