[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Is the list still working?

From: Jonathan S. Shapiro
Subject: Re: Is the list still working?
Date: Fri, 13 Jan 2006 20:29:56 -0500

On Fri, 2006-01-13 at 17:02 +0100, Bas Wijnen wrote:
> In that e-mail and the e-mails before it, I described a memory management
> system ...
> So the real question is: Do you think the proposal would be a good idea?  If
> not, and you want something which doesn't support self-paging instead, don't
> you think that's a severe limitation?
> > but I think we are about to send out a concrete design for pinning, and I
> > suggest that we should refocus our discussion around that.
> I look forward to it.


First, I apologize, but I am still unable to extract a clear proposal
from your email. It discusses several points at high level without
consideration of implementation, and some of the things that it appears
to describe do not (to me) appear to be implementable within the
restrictions imposed by other design constraints. In particular, our
continuing confusion about the MxN issue suggests that the design is
definitely not viable.

So: let me answer two of your questions specifically, and then let me
get the new proposal written down for discussion so that we can all look
at them together.

First, you ask (above) whether a design that does not support
self-paging is a severe limitation. My answer is: no, I absolutely do
NOT think that this is a severe limitation, for two reasons:

1 The term "self paging" is being used here as a code word covering
  a collection of imagined requirements. These need to be teased
  apart and examined individually. Until that is done, self paging is
  just a fuzzy comfort word, not a requirement. What we need to know
  is what the *real* requirements are and why.

2 There are fundamental problems in the interaction between certain
  aspects of self paging and the checkpoint subsystem. I would like
  to find a satisfactory compromise, but if I am forced to choose one,
  persistence is more important than self paging.

Second, you ask what the problem of O(MxN) storage is. You apparently
assume that this can be user-mode storage, but I think that it cannot.
That is the problem.

In the next note, I will try to tease apart the assumptions underneath
the concept "self paging" so that we can examine what is really needed.
On the coyotos-dev list, I will describe two mechanisms that Neal,
Marcus, and I have come up with in the last week. If you want to join
that thread, you have about 30 minutes to join that list at



reply via email to

[Prev in Thread] Current Thread [Next in Thread]