Subject: Re: Accounting for differences in physical memory performance
To: Ken Seefried <ken@seefried.com>
From: Jason R Thorpe <thorpej@wasabisystems.com>
List: tech-kern
Date: 05/22/2002 14:55:28
On Thu, May 16, 2002 at 02:27:17AM +0000, Ken Seefried wrote:

 > Is there a way, implimented or theoretical, that would maximize the 
 > performance of systems with different physical memory performance 
 > characteristics? 

Yes, NetBSD currently has a way to handle this (I added it a couple
of years ago).

What it does is break up the memory into multiple free lists, sorted in
order of descending priority.  The way this is used is that you would put
your fastest memory on freelist 0, next fastest on freelist 1, and so on.

The mvme68k port currently uses this facility for VME expansion memory.

It's not perfect (in particular, if a process gets stuck with a crappy
page, it could wind up using it for quite some time), but it seems to
be pretty effective at minimizing the bad effects.

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>