On Thu, 11 Nov 2004 16:35:50 -0500 (EST)
John Klos <john@klos.com> wrote:

> Did I miss something in this thread? Are you saying that there's some 
> reason to believe that this problem happens more with a different
> amount of memory? What is that based on?

Yes, I posted earlier today that the kernel panics go away when I add an
additional 128M of memory to the existing 128M. I believe that when
there is more active memory than physical memory some memory is not
being marked dirty, so the CPUs are losing cache coherency. With only
128M of RAM, a first time build of -current userland kernel panics in
less than an hour. With 256M it successfully built after 21 1/2 hours. I
could never get the active memory to exceed 256M, although Michael
suggest compiling Quake (which I haven't done yet).

tim