Message-ID: <qnxk7FApopbyEw+b@chocolat.foobar.co.uk>
Date: Wed, 23 Oct 1996 23:20:25 +0100
To: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
Cc: djgpp AT delorie DOT com
From: Paul Shirley <Paul AT chocolat DOT foobar DOT co DOT uk>
Subject: Re: Virutal memory problems.
In-Reply-To: <Pine.SUN.3.91.961023090200.6813I-100000@is>
MIME-Version: 1.0

In message <Pine DOT SUN DOT 3 DOT 91 DOT 961023090200 DOT 6813I-100000 AT is>, Eli Zaretskii
<eliz AT is DOT elta DOT co DOT il> writes
>
>On Mon, 21 Oct 1996, Paul Shirley wrote:
>> In one sense it really is wasted, the block is allocated and you won't
>> suddenly find it reused. However virtual memory paging will almost
>> certainly scavenge the unused pages if required, though frankly I prefer
>> to avoid the wastage in the first place.
>
>The problem is not the waste (AFAIK, it isn't wasted), but rather the 
>slow-down on the program once it starts paging.  Even if unused pages are 
>swapped out, the program really crawls, because many DPMI servers aren't 
>smart about virtual memory implementation.

In principle, a good VM manager will notice that the pages never get
touched and never do more than allocate the page descriptors. If that
happens then there's no loss as such and paging simply won't happen for
the unused blocks.

Of course that presupposes having a good VM system ;)
Things are a little more complicated with smaller blocks where paging
can't help. There you really will lose the memory (about 33% on average
with random block sizes).

-- 
Paul Shirley