Why is it not possible to combine a global replacement policy and a
fixed allocation policy?

Hello Sumavind,

The short answer is because those two methods have conflicting goals.
Fixed allocation implies the number of pages allocated for each
process is "fixed". Global page replacement implies the number of
pages allocated for the system is fixed and the number allocated to
each process can vary. Some operating systems use a hybrid solution
(minimum working set is fixed, maximum is adjustable).

For a good summary of Operating System page replacement policies check
out:

http://groups.google.com/groups?q=operating+system+fixed+allocation+global+%22page+replacement%22&hl=en&lr=&ie=UTF-8&scoring=d&selm=fa.icr6c2v.6hakhr%40ifi.uio.no&rnum=1
This is a newsgroup article titled "Understanding and measuring memory
usage on UNIX operating systems".  Look about 8-10 paragraphs down
where it describes working sets and page replacement algorithms. Later
it describes in more detail how Unix implements global page
replacement.

For more information, try a search with
  operating system fixed allocation global "page replacement"
in either www.google.com or groups.google.com (for newsgroup
articles). The web sites will provide a number of presentations or
class notes on the topic of page replacement and most describe both
global and fixed policies.

  --Maniac