NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/lang/pascal/4883

Text File | 1992-08-17 | 2.6 KB | 52 lines

Newsgroups: comp.lang.pascal
Path: sparky!uunet!cs.utexas.edu!torn!news.ccs.queensu.ca!mast.queensu.ca!dmurdoch
From: [e-mail] (Duncan Murdoch)
Subject: Re: Expanded memory and the heap...
Message-ID: <[e-mail]>
Lines: 40
Sender: [e-mail] (Netnews control)
Organization: Queen's University
References: <[e-mail]>
Date: Mon, 17 Aug 1992 13:26:23 GMT

In article <[e-mail]> [e-mail] (Falke Bruinsma) writes:
>
>I am writing a program that needs huge amounts of memory, at this moment
>around 700K. Basically it is one huge tree and I add new records to it
>by the following command NEW(ptr); However around the 600K (you guessed it)
>I get a heap overflow... Since I do have expanded memory I as wondering if
>someone can help me write a unit that makes the expanded memory an extension
>of the heap. So that the program accesses expanded memory whenever the heap
>is full.

It's not too difficult to put the EMS page frame into the heap, if 64K more
memory would be enough; Compuserve has a program that does this, and also
any other "high" memory that it can find.  That's usually not much more than
100K though.

If you really want full access to your EMS, you need some sort of paging
scheme.  I saw an ad once that a German company called OKSoft (sp?) had a
package that patched itself into the TP compiler and made this automatic,
but I've never heard anything more about that in the last year or so.

If you want to write it yourself, it's probably not practical to make the
paging transparent.  The best approach is probably to use "handles"; instead
of pointing to the thing you've allocated, they point to a record saying
where to find it.  You write procedures to dereference a handle by checking
the record and loading the thing if necessary.  It gets complicated if you
want to be able to dereference several handles at once, because they might
not all fit in memory at the same time.

A simpler approach which might well be good enough, is to keep your data on
a stream.  Use Seek, Read and Write every time you want to access or change
it.  Then you can have as much data as you have disk space, and if you've
got enough EMS to hold the whole dataset, you can load it into an EMS stream
and get the speed boost.  I've been using this method lately, and living
with the relatively slow access.  I use a TempStream from my Streams
unit; that automatically chooses between RAM, EMS, XMS and disk,
according to priorities you set.  Streams is available as garbo.uwasa.fi:/pc/
turbovis/stream13.zip by ftp.

Duncan Murdoch
[e-mail]