From: Tarik Nesh-Nash (tarikn_at_microsoft.com)
Date: Wed Jan 21 2004 - 15:04:49 PST
The MIT exokernel operating system is a thin layer between the hardware
and applications. Instead of providing application abstractions, like
traditional operating systems, exokernel intends to multiplex the
hardware. Application libraries will provide OS abstractions instead.
This can have a major advantage in performance and applications scope.
The exokernel separates protection from management and multiple OS can
then share the hardware resources. The applications manage what they
own (e.g. virtual memory, file system, etc)
The paper discusses the different layers needed to implement a complete
OS over the exokernel and discuss its performance results comparing to
mature UNIX operating systems. The performance is at least as good as
the traditional application in UNIX applications and it is much faster
when using specialized LibOSs (e.g. Cheetah web server).
This paper was a refresher of how much current applications are limited
due to constraints on the operating systems. As I m thinking of ways to
implement my distributed build process (my assignment project), the
idea of providing an OS implementation based on the exokernel seemed
very attractive as it will be aggressively optimized for performance.
The question that was not discussed on this paper is the cost. What is
the cost of implementing an OSLib? Is it worth having all this effort
spent to implement one application?
This archive was generated by hypermail 2.1.6 : Wed Jan 21 2004 - 15:04:36 PST