I've been following all of this VM system discussion, and thought I'd
add a couple of observations.  Also let me add that I'm coming at this
from a background where I know almost nothing of the internals of the
current VM system, so if I say something that doesn't make any sense,
I'd like to hear about it.

The discussion I've seen so far has listed two possible methods for
fixing the VM/IO cache problem.  One is to route the IO system through
the VM system (and its cache).  The other is to route the VM system
through the IO system (and its cache).

I want to mention that there is a third alternative that occurs to me
whenever I see this sort of problem (two modules with subsystems
having similar functionality):  why not create a completely new
abstraction to handle the problem?  Instead of talking about having a
VM cache or an IO cache, why not just have a single "cache" that both
the IO system and the VM system use?  

Anyway, since it looks like it will probably be a long time before the
VM system is overhauled and replaced, maybe the best solution really
is to just install the FreeBSD one.  I'd even be willing to help,
though I'm not the most qualified person.  And I certainly don't have
the time to do it on my own (unless we want to wait a couple of
years).

Todd Williamson