Subject: Re: newlock2 changes vs load
To: Zafer Aydogan <email@example.com>
From: Andrew Doran <firstname.lastname@example.org>
Date: 02/13/2007 02:06:14
On Tue, Feb 13, 2007 at 09:56:56AM +1100, Paul Ripke wrote:
> On Mon, Feb 12, 2007 at 08:31:27PM +0100, Zafer Aydogan wrote:
> > This also applies to LFS.
> > If I run LFS as a root file system, then my load is slightly above about
> > 1.12
> > Now with 4.99.10 my load idles at 3.12 (1+2)
> I'm not running -current yet, but I'm guessing we have another 1+
> kernel threads ticking over enough to add to the load average.
> I remember seeing this old post:
4.3BSD Net2 doesn't include the swapper in the load average calculation. I
also think we should exclude kernel threads given that the scheduler no
longer alters their priority. I think the test against PZERO in uvm_meter()
is intended to weed out LWPs that are sleeping interruptably. If so, that
one was broken before NetBSD came into being :-).
if (l->l_priority > PZERO || l->l_slptime > 1)
With both of these fixed I now get reasonable numbers for load average..