Better BestEffort : From Plato to Pope

Presentation given the 26th of October 2002 at the University of Waterloo.

Virtual memory management is one of the cornerstones of multiuser
operating systems. Most systems available today place all of the
policy in a monolithic virtual memory manager, VMM, isolated from the
rest of the system. Although secure and lightweight, users have no way
to communicate their anticipated memory needs and usage to the system
pager. As a result, the VMM can only implement a global paging policy
(typically, an approximation of LRU) which may be good on average but
is best for nobody.

With the port of Hurd to the L4 microkernel, this situation is being
readdressed. Due to its more distributed nature, a centralized
resource manager is not only more difficult to implement efficiently
but also contrary to the philosophy of the rest of the system. We are
currently exploring a model whereby each program is fully self-paged
and all compete for memory from a physical memory server. This talk
will first discuss how paging currently works in Mach and other
systems. An argument for an external paging policy will then be
presented followed by the requirements of such a design and the design
itself.

A recording is available at:
http://www.csclub.uwaterloo.ca/~ja2morri/neal.ogg

Copyright (c) 2002 Neal H. Walfield

Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.1 or
any later version published by the Free Software Foundation; with no
Invariant Sections and with no Front-Cover Texts, and with no
Back-Cover Texts.  A copy of the license is included in the file
COPYRIGHT.