Subject: Re: VM problems
To: Bill Studenmund <firstname.lastname@example.org>
From: Mark F Willey <email@example.com>
Date: 07/28/1995 23:50:21
"In a previous message, Bill Studenmund wrote:"
> On Fri, 28 Jul 1995, Mark F Willey wrote:
> > "In a previous message, Bill Studenmund wrote:"
> > >
> > > Could a process-memory limit (ULIMIT ?) be set lower than 16 Meg (or even
> > > to exactly 16 Meg)? To test this out, try running 10 of the 8 meg
> > > versions. :-)
> > AFAIK, NetBSD does something called "copy-on-write", which basically means
> > that X copies of a program won't take up (size * X) amount of memory.
> > They will just take up (size) of memory.
> > (plus overhead, maybe) If the program writes to the image, then, I think
> > it will "copy" it, and split that process's text space off. I'm sure
> > someone else here could explain it better! ;-)
> Ok, but won't they each have a copy of the data each one malloc's? The
> program would allocate x meg of data, and then it wrote to the
> data it'd allocated. Thus even if the system had a delayed-allocation
> method, it'd have to allocate the memory from vm.
Yes. The data they use would be linear. The binary image is the only part
that is not. So your idea might work.
Maybe I'm not thinking right on what kind of program this is.