[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference hydra::amiga_v1

Title:	AMIGA NOTES
Notice:	Join us in the NEW conference - HYDRA::AMIGA_V2
Moderator:	HYDRA::MOORE

Created:	Sat Apr 26 1986
Last Modified:	Wed Feb 05 1992
Last Successful Update:	Fri Jun 06 1997
Number of topics:	5378
Total number of notes:	38326

2801.0. "Memory Defragmentation" by KYOA::MIANO (O.K. so who cares about the METS?) Mon Aug 07 1989 00:36

I am working on an application written in C that uses a lot of stack
storage.  I'm not sure how much right now but it may be in the 1-200,000
range.  Unfortunately the Lattice Debugger has a bug that allows you to
only access 1/4 of your stack.  This means that while debugging I need
to set the stack from 4-800,000.  This would be all right since I have
a 2500 with 3 Mbs except for one problem: memory fragmentation.  With
the stack set to 800,000 often a process gets completely stuck.  You
can't even run the stack program even though there are 2.5Mbs of free
memory. 

My questions are:

1) Is is possible to deframent the Amiga's memory.  Is it possible the
the OS may break the memory into chunks that may be contigouos but the OS
cannot combine adjacemt chuncks?
2) Can anyone recommend a good tool for memory defragmentation?

John

T.R	Title	User	Personal Name	Date	Lines
2801.1	Exec does defragmentation	TLE::RMEYERS	Randy Meyers	`Mon Aug 07 1989 01:37`	20
	Re: .0 Sorry, the Exec already does memory defragmentation when memory is deallocated. The cause of fragmentation is small bits of memory that really are allocated and prevent regions from coalescing. (Well, there is one type of unnecessary memory fragmentation. The autoconfig code will not coalesce memory from two separate memory boards even if the memory does really form a big contiguous chuck of your address space. So if you have separate memory boards, or you have a 8 Meg board with 6 Meg installed, you could experience this type of fragmentation. ASDG wrote a public domain utility to remedy fragmentation due to autoconfig.) About the only thing you can do is make a boot floppy with your application on it, and try to get into your program as soon as possible. Don't use any ram disks. Don't do any assigns. Don't load the Workbench. Don't run any screen blankers or "pop up" programs. Don't (or try not to) do anything that uses memory before running your program.
2801.2	Do your own alloc()	FROCKY::BALZER	Christian Balzer DTN:785-1029	`Mon Aug 07 1989 05:13`	9
	There was a fine article about this topic by one of the Amiga engineers on UseNet quite a while ago. I'll check if I can locate it. What it basically boils down to is to have your program allocate a sufficiently large chunk of memory and do all the tiny allocs with a custum alloc function from this pool. If you exit the application fragmentation will be minimal. Regards, <CB>
2801.3	Right answer, different question?	TLE::RMEYERS	Randy Meyers	`Mon Aug 07 1989 18:01`	33
	Re: .2 I think you've answered a different question, Christian. The problem is that all the other programs in the system have fragmented memory, and so there isn't memory enough for the user written application to start. The author of .0 was looking for something to do to the other programs in the system to keep them from leaving memory fragmented. > What it basically boils down to is to have your program allocate > a sufficiently large chunk of memory and do all the tiny allocs > with a custum alloc function from this pool. I think the problem here is that the Exec will gladly carve memory up into tiny chunks for tiny allocation requests. As long as those chunks remain allocated, the system memory pool will look like Swiss cheese. However, then those chucks are deallocated, the Exec will coalesce them back together. The scheme you've described allows an application that allocates lots of little chunks to run at the same time as an application that needs a few very big chunks. > If you exit the application fragmentation will be minimal. As I said before, if you deallocate the memory, it get coalesced. The problem comes if you don't deallocate it, or somebody else needs to run at the same time with you. > There was a fine article about this topic by one of the Amiga engineers > on UseNet quite a while ago. I'll check if I can locate it. Please do. I would be interesting reading.