the current implementation of a process address space is a linked list.
applications like libefence won't scale well until we switch to using
a better data structure to represent an address space.

anyone want to write some splay trees?  :-)

-Chuck


On Tue, Feb 15, 2000 at 10:42:13PM +1100, Julian Assange wrote:
> 
> I noticed when running libefence on a heavy small object mallocer, increasing
> slowdown for each mmap. libefence maps pages like so:
> 
> 
>         OXOXOXOXOXOXOXOXOXOX
> 
> for each malloc, inorder to catch off region reads and writes. If whatever pmap/uvm
> mechanism being applied only has linear page indexing, this would account for
> the observed behavior. Is this the case?
> 
> Jason? Chuck?