Subject: Re: bufq: readprio vs priocscan dbench results
To: Juan RP <email@example.com>
From: Thor Lancelot Simon <firstname.lastname@example.org>
Date: 10/11/2005 08:29:27
On Tue, Oct 11, 2005 at 10:26:17AM +0200, Juan RP wrote:
> On Mon, 10 Oct 2005 13:15:08 +0200
> Juan RP <email@example.com> wrote:
> > Hi,
> > I've been running dbench to benchmark our current bufq
> > strategies and here are the results:
> More results, this time with softdep enabled/disabled:
These are very interesting. But I am curious -- it doesn't seem like
you retested with the 6-queue priocscan configuration that gave the best
performance without softdep in your earlier test?
Also, some numbers for average request latency might be very revealing.
I think you could get these with postmark (the NetApp benchmark) though
you may need to tweak its tunables some.
This is all very good work. We've needed someone to focus on this and
do the laborious work of performance characterization for some time.