Re: Getting Started with Hurd-L4

[Marcus Brinkmann]

At Tue, 26 Oct 2004 13:01:22 +0100,
Sam Mason <address@hidden> wrote:
> 
> One thing I did just think of is something that Linux calls
> "over-commit".  Not sure if this system will be affected by it, but
> when I read the above paragraph that's what immediatly sprang into
> mind.  If memory pages are given out on a demand only basis, are tasks
> going to be affected by physmem making promises that it can't keep?

Neal will have to clarify this (see the other thread I started
yesterday, where I pose a similar question in my reply that I just
sent).

There is an easy answer to it, and a hard one.

The easy: If you over-commit (have more frames than guaranteed), the
remaining frames are "extra" and physmem might ask them back at any
time (you can't even swap them out).  _Which_ of your frames are
extra, and which are guaranteed, is up to you.

The hard one: How to negotiate the number of guaranteed frames per
task?  Ie, how to distribute the available physical memory fairly and
squarely in a living system with changing number of users and tasks?
This is the thing that's still in the dark for me.

Thanks,
Marcus