[Top][All Lists]

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

Re: currency based virtual memory

From: Matthew Sackman
Subject: Re: currency based virtual memory
Date: Thu, 7 Nov 2002 13:23:38 +0000
User-agent: Mutt/1.4i

On Sat, Nov 02, 2002 at 11:20:25AM -0500, Neal H. Walfield wrote:
> > 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.
> With respect to the allocation memory allocation policy, I have not
> gotten far enough in my research to either say this is or this is not
> my model: the market based approach is only an idea, but one that I
> think would certainly be interesting to explore in more depth.

Possibly interesting, but I would personally not like it as it will
create a heirarchy and that somewhat goes against 'free as in freedom'
for the following reason:

Suppose x people are logged into one machine and are using it with
similar workloads. Then, say one person wants to compile an
application. With the economic model it would *appear*[0] that what
would happen is that in order to optimise the global system, the system
would delay that compilation as being too memory exhaustive and too cpu
intensive. Thus as a result the system becomes a 'do anything resource
intensive if and only if no one else is logged in', which to me, does
not seem to be a 'free' system.

The current systems would, I imagine, go the other way - do the
compilation, and everyone suffers! Still, the possibility exists that
everyone else that was logged in, had actually gone for their tea break.
If the system delays the compilation then not only did the programmer
suffer a delay unnecessarily, but the other users were basically, though
unknowingly, performing a DoS attack on the programmer by denying them
the possibility to compile.

If optimisation of the global and local system (as defined in [1]) can
be achieved through memory management controlled by a system based on an
economic model then I believe it needs to be very closely linked to a
user's 'activity' and their cpu usage - the system should not prefer
people using X over console simply because X takes up more cpu cycles.



[0] Please note I have not fully read all the documents cited, so I may
be reading falsities into the economic model.
[1] http://citeseer.nj.nec.com/367323.html
Matthew Sackman

BOFH excuse #206:
Police are examining all internet packets in the search for a 

reply via email to

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