[Top][All Lists]

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

Re: currency based virtual memory

From: Niels Möller
Subject: Re: currency based virtual memory
Date: 02 Nov 2002 13:53:31 +0100
User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2

Sunnanvind Fenderson <address@hidden> writes:

> Recently I came across a discussion of using a market-like or
> currency-based system for virtual memory. I'm specifically replying to
> what Niels wrote at
> http://mail.gnu.org/pipermail/l4-hurd/2002-October/000709.html .

When replying to old messages, please insert an in-reply-to: header
referring to the article in question. It makes it a lot easier for
people who are subscribed to the list (and uses decent mail clients)
to review the message you're replying to.

> Niels, I think you're overextending the metaphor somewhat. The
> "currency system" could simply be a system of users allocating *what's
> available at the moment*, meaning no saving, no speculation and no
> interest.

If we buy into the theory that economic models is a good way to do
distributed resource optimization (I don't really know the theory, but
I think the trick is to formulate an optimization problem with
constraints, and then interpret the lagrange multipliers in the
solution as prices), then interests seem quite natural.

The economics isn't just a "metaphor", it's a method of doing
distributed optimization.

The point of introducing interest, I think, is that if I can choose to
run a memory intensive job today or next week, it's better for
everybody else if I do it next week (perhaps it turns out that the job
is not needed anymore, or perhaps there will be a memory upgrade in
the weekend). Introducing interest on memory which I don't use
(effectively lending it to other processes and users) creates an
incentive for me to postpone that job.

You can also compare it to the process priority in ordinary unix
scheduling. Basically, processes that sleep most of the time get
higher priority than processes that want to consume cpu continuously.
One could think about that as an interest on cputime the process
didn't use.

As for my speculation speculation, I agree that's a little
far-fetched. One can't really save memory for later, either you use it
or it's wasted. However, as I understand Neal's model, processes buy
contracts for memory for some period of time, and if it's possible to
buy longer-term contracts, that opens up possibilities for speculation
and interest loans, even if the central memory server doesn't
implement interests.


reply via email to

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