Re: making paging decisions

[Marcus Brinkmann]

At Tue, 26 Oct 2004 09:23:50 +0100,
Neal H. Walfield wrote:
> Physmem only reclaims extra pages.  Are you going to all of this
> trouble for extra pages?

That is a good question.  Maybe I am exaggerating the problem due to
lack of confidence and oversight.

But then, you have not said much at all yet about how the number of
guaranteed frames is determined, and how it can (if at all) change
over the run time of a task (and what happens if there is memory
pressure).

If we give too little guaranteed frames, then all (or some) tasks in
the system will be hopelessly swapping.

If we give too much, we may run out of memory if many tasks are
started, and we will get into memory pressure that can only be
relieved by revoking guaranteed frames (ie canceling the contract and
renegotiating).

My personal feel is that we can choose to ignore this now, but at some
point people will expect to be able to handle the number of guaranteed
frames in a more dynamic way than we have seriously considered so far.
If I look at the processes running in my system, most of them use
under 1MB RSS, or just a few MBs.  But there are also significant
spikes (hello emacs, which apparently uses ~110MB RSS, yuck!).  Heck,
even my gnome-terminal uses 10MB RSS.

How do we solve the problem of a task needing more guaranteed frames
than we can give to everyone?  I don't want emacs to be constantly
swapping if there is plenty of memory left in the system.

Thanks,
Marcus