Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!purdue!decwrl!mejac!gryphon!amy!hansel
From: hansel@amy.uucp (Steve Hansel)
Newsgroups: comp.sys.amiga.tech
Subject: Re:Re: free memory defragmentation
Message-ID: <7354@amy.uucp>
Date: 22 Oct 88 18:17:28 GMT
References: <9887@cup.portal.com>
Reply-To: hansel@amy.uucp
Organization: Hansel-Graphics
Lines: 36

> I am pretty sure that this IS done with a normal FreeMem call since
> the nodes in the memory free list ARE sorted and we get big chunks
> of free memory if we free small adjacent blocks before.

> The problem of fragmentation is another one. It is not small adjacent
> blocks that aren't concatenated BUT small NOT adjacent blocks that
> can't be concatenated.

How about allocating 'small' chunks from the top of memory downward and
'large' chunks from the bottom of memory upward.  This would mean that,
unless your memory was very full,the fragmentation would only occur
between small chunks.

I bet this could be done with a minimum of overhead.  The biggest impact
would probably be that the free list would be comprised of 12 byte blocks
because it would be doubly linked.  This shouldn't be a problem because there
was never any guarantee on the minimum size of alloced blocks.

Also future releases of the O.S. could have optional MEMF_PERM (or
MEMF_LONGTERM) flag that would force the OS to allocate from the small
chunk area (Top).  This flag would be used by Clock programs, system
monitors, shells, etc, when they AllocMemed because they tend to keep their
memory for long periods of time.

I think this approach is much easier than allocing larger chunks to hold
smaller chunks, or using pointer handles.

   Just a thought


         Steve Hansel

                  gryphon.cts.com!amy!hansel
                  oberon.usc.edu!amy!hansel
                  rutgers!amy!hansel