Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!ukma!rutgers!cbmvax!daveb
From: daveb@cbmvax.UUCP (Dave Berezowski)
Newsgroups: comp.sys.amiga.tech
Subject: Re: free memory defragmentation
Message-ID: <5005@cbmvax.UUCP>
Date: 17 Oct 88 02:36:53 GMT
References: <9887@cup.portal.com> <3009@amiga.UUCP> <10003@cup.portal.com>
Reply-To: daveb@cbmvax.UUCP (Dave Berezowski)
Organization: Commodore Technology, West Chester, PA
Lines: 35

In article <10003@cup.portal.com> dan-hankins@cup.portal.com writes:
>In article <3009@amiga.UUCP> jimm@amiga.UUCP (Jim Mackraz) writes:
>
>>By the way: memory fragmentation on the Amiga occurs because there are
>>actually allocated sections of memory between the freed chunks.  There's
>>been a lot of discussion on this already.
>>
>>Note that you can't concatenate two non-adjacent chunks.
>
>Ah.  So what we need is an MMU (i.e. virtual memory).
>
>Hmmm.  What would happen if the ALLOCATE and DEALLOCATE functions were
>re-written in such a way that all memory would be allocated in 1k chunks?
>Then successive calls to ALLOCATE would check to see if part of an
>already-allocated chunk was available for allocation, and use that if it
>could.
>
>That is, first call (100 bytes) results in a 1k chunk being allocated. 
>Second call results in a second pointer into the same 1k chunk.  This might
>reduce the problem you spoke of without as much need to resort to an MMU.
>
>If we could keep track of all pointers to a memory chunk, we could
>rearrange them no problem.  But then the Amiga OS isn't quite *that*
>object-oriented.
>
	Well you could keep ptrs to all your variables in one structure
(sort of like a freelist) and periodically call a system routine
like 'UnFragMemory' which would attempt to un-fragged memory by re-working
all your variables (ie. all your memory use); copying the data as it
needed to.

	Of course this scheme relies on the application writer; and we
all know how dependable they are! :^)

	Regards, David Berezowski