Subject: Re: Virtual Memory Subsystem
To: None <>
From: John S. Dyson <>
List: port-i386
Date: 11/26/1996 18:11:22
> > 
> > Well, there are several cases, the most common is actually just renaming
> > pages which is cheap.  There is however the "expensive" case which occurs
> > very seldom, and that requires pagein alright.  Actually BSD/OS skips this
> > case, apparently just the "cheap" collapsing operation suffices for them.
> Now I can see why the BSD/OS fix doesn't work in my BSD/OS machines at all.
> That case probably is not as rare as you think.
We went through a couple of "partial fixes" on FreeBSD, and finally gave in
and did a complete implementation.  Ours could *definitely* be cleaned up,
but it is hard to motivate myself to fix something further that already works.
It had been a thorn for us for long enough.

Every time we thought that we tweaked it to make another leak go away, we
found another circumstance that would cause it to come back.  The best
thing to do is to just fix it entirely.