How does the BSD server implement sync (how does it find all the dirty
pages)?

Would LFS have problems in its buffer management if VM owns the pages?

What's the best (most efficient) way to copy the heap and stack
segments for a fork()?  There isn't any way to do COW on
externally-managed memory objects, right?

What does DLB (and other OSF folks) run?  OSF/1?  Mach 3.0 (using what
server)?

Should the Sprite server try to keep detailed records about what
portions of the client's address space are mapped by what Sprite
objects?  (My inclination is to say "no", because there's no guarantee
you'll have the right information.  However, this complicates
VmAddrParse.)

Any thoughts on how to handle address exceptions in server (e.g.,
short read from FS)?