Re: currency based virtual memory

[Niels Möller]

address@hidden (Neal H. Walfield) writes:

> Well, it should not swap everything out.  But if you really way to,
> then I imagine you would have to ask someone else to wake you up.

That brings up another issue I wanted to ask about before. What if a
process doesn't give back RAM to the system when the memory manager
asks it to? Your slides say that the memory manager can decide to
revoke all mappings (which is more or less the same thing as killing
the process, right?).

I find it a little scary that processes that miss a memory-return
deadline will be killed with no mercy. Will they even get the chance
to dump core?

A better alternative might be to swap out the process. And here I'm
talking about the old fashiond whole-process swapping, not the modern
paging variant.

Is there any way for a (privileged) memory manager to do that? I think
the steps needed to swap out a process are to (i) suspend all its
threads, (ii) copy all it's pages to disk (if the aren't on disk
already), and (iii) reclaim all its pages. Somebody still needs to
keep some record enough of it's state (threads and mappings, I guess),
so that it can be revived later.

This old-fashined mechanism may touch with process persistance and
migration, but I won't pretend to be familiar with at.

/Niels