Subject: Re: Virtual Memory Subsystem
To: Jukka Marin <>
From: John S. Dyson <>
List: port-i386
Date: 11/26/1996 08:16:50
> > I looked at the OpenBSD "fix", and was a "solution" that we specifically
> > decided not to implement on FreeBSD 3yrs ago.  It is better than nothing
> > though.  It is impossible to fix the problem unless at least a solution
> > of the type that OpenBSD has implemented is put into place.  I disagree
> > with the OpenBSD "fix", because it is sub optimal, but is certainly easier
> > than the complete (and complex) fix that FreeBSD has implemented.
> Well, we should also get the unified buffer cache (is that what you call it?)
> sometime.. so it might be worthwile to do both changes at the same time(?).
> I'm not trying to help in postponing the fixes, though.. :)
The collapse fix is orthogonal to the merged cache.  If I was a product
manager deciding which was more important (evaluating risk/rewards), I
would go for the following order:

1)	Collapse fix
2)	Pageout perf
3)	Merged cache

The merged cache helps alot of things, but frankly only helps a little
(IMO.)  If someone has an eccentric workload, I believe that they can
increase the size of the buffer cache manually...  I think that users
would like a reliable OS rather than extracting all the perf from a
system.  Merged cache mods are more likely to destablize NetBSD than
collapse fixes or pageout perf mods.

The collapse fix is difficult mostly because it requires mods to the
swap pager to allow it's internal metadata to be manipulated on a
page-by-page basis.