From: Ankur Rawat \(Excell Data Corporation\) (a-arawat_at_microsoft.com)
Date: Wed Feb 11 2004 - 15:56:25 PST
This paper breaks new ground in developing a OS Virtual memory
management system using paging. The new approach is in developing a
system by seperating the notion that good VMM can only be achieved by
tight coupling between hardware and OS VMM system.
I do not know if this was the first approach to develop a mostly
hardware independent VMM system, as I have always read that good VMM
system is not possible without close hardware support.
The VMM system is built in the Mach OS developed at CMU. It has been
ported to multiple hardware architectures. The system does not make any
assumptions about the H/W but only expects it to handle and recover from
page faults. It is a message based system and tries to achieve its
performance by integrating whole file transfers and whole address spaces
with message sending.
The system is split into 2 parts:
1. Machine dependent portion (consists of only 1 module and a
header file): It implements those operations necessary to manage H/W
data structures
2. Machine independent portion
The porting experiences suggest that it did not take a lot of effort to
port Mach VM to different H/W architectures.
Issues encountered while implementing the system on different H/W
architectures:
1. Keeping page tables in Physical memory complicates the machine
dependent code
2. One valid mapping of each physical page makes it impossible to
share pages without faulting
This archive was generated by hypermail 2.1.6 : Wed Feb 11 2004 - 15:56:39 PST