In article <199802280859.JAA24589@father.ludd.luth.se> you write:
> Yes, I have fixed support for it, it will go in later this weekend.
> It decreased the rc run-time on my MVII with 10 seconds, from 1m50s
> to 1m40s :-) This may also be because of much fewer calls to the 
> pmap functions in UVM; a function calls takes 20-30uS on a MVII.

Cool! And that's just with the current UVM.  Much of the drive behind
UVM, as I understand it, is to eventually offer page loaning (for
networking, etc) and a merged buffer cache, which should make things
even better.

> up and running. I don't remember any numbers anymore, but compared to
> 4.3 Reno the VM system was a great improvement; especially when creating
> new processes, the Mach VM is also much better in caching. Probably the

And UVM improves even more... All the more reason to work on fixing
NetBSD/vax's weak points instead of throwing out the baby with the
bathwater.