Date: Fri, 16 Sep 94 15:01:00 EDT
From: peprbv AT cfa0 DOT harvard DOT edu (Bob Babcock)
To: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: GO32 paging algorithm
Reply-To: babcock AT cfa DOT harvard DOT edu

> The paging algorithm used to be round-robin.  Pages would get selected
> in order, regardless of activity.  The current algorithm is supposed
> to be LRU, with protection around IP and SS and the page being paged
> in.  However, be aware that some access patterns will cause LRU to be
> worse than any other paging algorithm.

Does LRU stand for Least Recently Used, and if so, does that mean that the
page which is swapped out is one which hasn't been swapped out recently?  It
would seem that what you really want to swap out is a page which hasn't been
used recently, but I don't think the 386 keeps any statistics on that.

I'm afraid my intuition fails at guessing what paging algorithm would be best
for sorting and merging.  Would it be reasonable to have a GO32 (Turbo
assist?)  call to select the paging algorithm?  I scanned very quickly
through paging.c, and it isn't obvious whether this would be a small change
or a big one.  I also wonder what algorithms are used in other environments?
I suppose I should run my benchmark under QDPMI, OS/2 and Windows.