Spring CSE590S
Announcements
From: schwerin@cs.washington.edu
Subject: Minix 3 590S Web Page
Date: April 3, 2006 3:42:57 PM PDT
To: bershad@cs.washington.edu
The short of it is that Minix-3 runs fine in VMWare these days, using an image you can download off the web.
Link it to here
Reading Schedule
- April 5. Andrew Schwerin.
A Lightweight Method for Building Reliable Operating Systems Despite Unreliable Device
Drivers>
. This is the Minix 3 paper.
- April 12. Brian Van Essen.
On
Microkernel Construction The L4 Architecture.
- April 19. Towards
Microkernel Construction A rebuttal against performance arguments against microkernels.
- April 26. Preventing
Denial of Service Attacks on microkernel for WebOSsThe title says it all.
- May 3. Day off
- May 10. Paul Gauthier. OSDI Submission: Case for a new search
architecture.
- May 17. Harsha Madyastha. OSDI Submission: iPlane: An Information
Plane for Distributed Services
- May 24. TBD
Tasks
Overview
We're going to use this term to learn more about two contemporary kernel systems. Motivation is:
- expand our toolbase for doing interesting systems work (now we know about linux, xen, and a little
NT)
- answer a couple of basic questions about the systems
- develop enough material to potentially write an "experiences" report
- get the students' hands dirty
The two systems are:
Minix3
L4
Both are microkernels that purport to run real applications. (I've booted and played around with minix; not yet
with l4)
We want to form two teams, one connected to each system. We want the teams to get educated on the internals of
the system, both from the reading papers side and the reading code side. We'd like each team to propose and
agree upon one or more substantial pieces of software to build on top of the system. I recommend:
a router (click)
a caching web proxy
Here's what we want to accomplish for the quarter:
- bring up the systems
- bring up the applications
- evaluate the systems and applications along at least these two dimensions:
- performance
+ footprint ("how small can you get?")
+ speed
- reliability ("can you get it up and keep it up?')
+ robustness in steady state
+ robustness in overload
+ robustness against random user attack (this is Bart Miller's syscall flooding)
+ evaluate against simulated random kernel and user errors (Nooks-style attack)
As baseline comparison, we do the same evaluation against the apps in linux.
Output is a three way comparative evaluation.