This is an attempt to try to solve the performance and CPU usage
problems in vm subsystem.

During the development of this patch I have found some interesting
things:

* Most read pages will end up as Referenced even if the PG_referenced
  bit is cleared when inserted into LRU - this is probably due to the
  pages being read ahead, and thus later referred...

Improvements/bugs in patch:

* does not handle age wrap of really old pages.

* does not use reused pointer.

* could use another counter (incremented at use) or
  mechanism instead (function that ages all pages at once).

* I am not sure how it will handle mmap002, forgot to run it before
  connecting to internet...

I am sending it as is since I will be away this weekend...

/RogerL

--
Home page:
  http://www.norran.net/nra02596/