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.

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.