memory allocation

[Robert Belanger]

Hi everybody,

	We are in the process of speeding up memory allocation. Because we
have a CPU with a linear address space, but no (easy) access to disk. We have
four(4) megabytes of DRAM available. What we want to do is avoid any system
in order to speed up memory allocation and also reduce the overhead associated
with memory allocation.

	We tought of several scheme

	1)	we could allocate a long block of memory and manage it 
	ourselves (by maintaining headers to our data blocks) but two problems
	arises 1) when we free a block then we have to scan the whole area in
	order to see if we could merge two blocks together. 2) scanning the
	list (on other system with swap space) induce a lot of disk swapping.

	2)	we could allocate a long block of memory and also create two
	lists in memory (one for free blocks the other for used ones). problems
	1) when somebody frees up a list we have to maintain in the block (or
	just before) where in our used list is the header to refers to it.
	2) it looks awfully complex to maintain a stable relation between the
	lists and the actual data.

So, I am asking your help, what do you think? do you know (or think) of any
other alternative?

Thank you very much
#if enough_interest
I post a summary of answers
#endif
===============================================================================
Philips Electronics Ltd.			     	     Robert L. Belanger
600 Frederick Philips Blvd		     	     Advanced Development Group
St-Laurent, Quebec, CANADA					   ECHO project
H4M 2S9	(514)-744-8200-2495		  		Eternity is very long!!
			        	  	       Especially in the end!!!
E-MAIL: philabs!micomvax!belanger

"Je me laissai fondre comme un suppositoire dans le derriere du
'Moonshine Bowling'", 
	Francois, Au clair de la lune.
===============================================================================