From:	sandmann AT clio DOT rice DOT edu (Charles Sandmann)
Message-Id:	<9607291335.AA14001@clio.rice.edu>
Subject:	Re: BIOS swap file
To:	djgpp-workers AT delorie DOT com (DJGPP developers)
Date:	Mon, 29 Jul 1996 08:35:22 -0600 (CDT)
In-Reply-To:	<Pine.SUN.3.91.960729083110.25853F-100000@is> from "Eli Zaretskii" at Jul 29, 96 08:36:42 am

> > What about a bios swap file (like windows) or partition? Sure, it limits swap
> > space to a fixed amout, but it should be faster than dos.
> Are you sure it will be significantly faster?  AFAIK, the BIOS is one of 
> the causes of the slow real-mode disk I/O, because it ties the processor 

I'm not going to do it - the reasons being:
1) Page faulting is dominated by the seek time on the disk.  There is nothing
   in DOS does which will be even close to the rotational latency.
2) The cost/benefit is poor.  With memory in the $6/Mb range, and the number
   of people who don't page at all currently, and the almost neglible 
   improvement in performance - It just doesn't make sense.

I did consider using the windows's permanent swap file if it was specified,
since it is known contiguous.  I also thought about letting you specify
an existing file, not deleting it on exit, so you could make a contig
place on disk for better performance.  Then the arguments above convinced
me that improving performance of paging is probably not cost effective.
Sorry.  

- Raw text -