idea for running out of RAM

[John Gilmore]

> I agree with Albert's proposal - Newcomers to the Wellington test team
> open too many apps all the time - and render their machines unusable
> through memory pressure. From that experience, I like the idea of
> adding a bit of metadata that hints the mem footprint, and teaching
> sugar to prevent users from starting new apps if less memory than
> stated is available.

We can slap a crutch like that onto the system, but we actually built
and shipped a release (703) that did let you run three or four
activities without serious trouble.  That release is in the field, and
people will expect to still be able to run three or four activities
after upgrading to the new release.  In theory, releases get better
rather than worse.

Our old release also took quick corrective action when it ran out of
memory -- it rapidly killed off some program, often the largest one.
Our new release doesn't do that, and nobody knows why.  Instead it
just gets really slow and groggy.  It seems to be a kernel change, but
nobody's bisected it.  See http://dev.laptop.org/ticket/8316, for
example.

Rather than putting work into crutches, let's fix the real problems
that we already know we have.

	John

PS: We don't even know the mem footprint of our activities.  It all
gets jumbled up by Sugar and Security and other changes.  Nobody
thinks the numbers in "top" are useful, nobody has any better way to
measure the "mem footprint" of an activity.  So even assuming that you
added a field to every activity, what values would you put in those
fields?