Re: Another question about self-paging

[Jonathan S. Shapiro]

On Mon, 2005-10-31 at 13:29 +0200, Paavo Parkkinen wrote:
> On Sun, 30.10.2005 at 16:55 -0500, Jonathan S. Shapiro wrote:
> > There is a meta-level problem with self-paging. It is predicated on the
> > assumption that one application should be informed about pressures
> > resulting from the behavior of other applications.
> 
> Why is this a problem?

Because any story about security begins with the goal that for any two
process A and B, it should be possible to prevent communication from A
to B. If B is asked to reduce its working set in response to an action
by A, then communication has occurred.

This is a surprisingly high bandwidth channel.

We will not, of course, achieve the isolation goal perfectly in any real
system, but it would be a grave error to architect in a feature that
makes any meaningful separation impossible from the beginning.

shap