cs591: Networks and Distributed Systems

Previous comprehensive exams: Spring 1997, Fall 1997, Fall 1998, Fall 1999, Fall 2000, Spring 2001, Fall 2001, Spring 2003, Fall 2003, Spring 2004, Fall 2004, Spring 2005, Fall 2005, Spring 2006, Fall 2006, Spring 2007, and Fall 2007.

Some solution files: Fall 2003, and Spring 2005.

A studying guide for the systems portion of the comprehensive exam by Prof. Patrick Bridges.

We covered this in class on August 30th. Rosenblum and Ousterhout proposed the log-structured file system in The Design and Implementation of a Log-Structured File System.

In Security and log structured file systems the author points out that it is nearly impossible to overwrite file data in a log-structured file system. While overwriting file data in a standard file system is not a secure or reliable way of destroying data, a log-structured file system makes it impossible.

Jambor et. al. describe a Linux implementation of a log-structured file system in Implementation of a {Linux} Log-Structured File System with a Garbage Collector.

Wikipedia has a nice short article about log-structured and another about journaling file systems.

Sarita Adve provides a very nice tutorial on shared memory consistency models in Shared Memory Consistency Models: A Tutorial.

Online, Joe Pfeiffer from NMSU has some nice class notes. Peter Druschel's class notes are also good and put consistency into a broader context.

The paper A Unified Theory of Shared Memory Consistency by Robert C. Steinke and Gary J. Nutt is quite theoretical but may help with understanding the relationships among the various consistency models.

I post the paper about slow memory here because it is referenced by the main papers on consistency models and it answers the question about physical exclusion that came up on the mailing list.

A good companion to A Theory of Memory Models is Threads Cannot Be Implemented as a Library.

Also, be sure to look at the Java memory model description which explains what memory models are in the context of a well-known programming language.

The examples in the paper are expressed in Java code at this web site.

On September 27 we discussed the BSD file system in class and briefly covered soft updates as an alternative to a log or journal on disk. The basic structure of the Unix (BSD) file system is explained in A Fast File System for UNIX. In Extent-like Performance from a UNIX File System the authors discuss a variety of improvements that can make a file system faster. McKusick gives a nice summary of the changes to the Unix file system over time in A Brief History of the BSD Fast File System. The idea of soft updates is introduced in Soft Updates: A Technique for Eliminating Most Synchronous Writes in the Fast Filesystem.

Lamport explains why he wrote Paxos Made Simple here. The story behind the publication of the original paper, published with an eight-year delay, is here.

Butler Lampson gave a talk about Paxos on the occasion of celebrating Leslie Lamport's 60th birthday. The write-up is here. Not necessarily easy reading...

Peterson's algorithm is one of the simplest to guarantee mutual exclusion. This web site has a nice description of the individual elements of the algorithms. By building the algorithm from scratch, it shows the purpose of each piece of the algorithm.

In addition to The Byzantine Generals Problem, this web page. gives a nice explanation of the algorithm. The history of achieving consensus in a distributed system is concisely explained in A brief history of Consensus, 2PC and Transaction Commit by Mark McKeown.

Wikipedia has good explanations for the Two-Phase Commit Protocol, the Three-Phase Commit Protocol. the ACID properties, and the Two Generals' Problem.

The unpublished paper A Survey of Distributed File Systems by Fabio Kon gives a nice overview of several distributed file systems and compares some of the main features of a distributed file system.

The technical report Prefix Tables: A Simple Mechanism for Locating Files in a Distributed System by Welch and Outserhout explains prefix tables in the Sprite network file system.

While reading Xen and the Art of Virtualization, you may want to visit this web page, which has a list of papers the Xen developers recommend to come up to speed on the topic. (Thanks Ricardo.)

The May 2005 issue of IEEE Computer had a feature on virtualization with very readable articles: The Architecture of Virtual Machines, Virtual Machine Monitors: Current Technology and Future Trends, Intel Virtualization Technology, and Rethinking the Design of Virtual Machine Monitors.

We did not really talk about distributed OSes in class, but if you would like to catch up, the paper Distributed Operating Systems by Andrew Tanenbaum and Robbert Van Renesse would be a good start.

The classic Peter Denning paper Virtual Memory gives a good introduction to the topic.

In Before Memory was Virtual Peter Denning describes the evolution of virtual memory and the problems it solves. A very nice read with a little bit of history woven in.

Page coloring is an important concept and shows up on the comps once in a while. The paper The Effect of Page Allocation on Caches describes the topic.

It is interesting to learn how memory management units work and implement the principles of paging and virtual memory. The paper Microprocessor Memory Management Units provides a nice introduction and then compares various MMUs.

This is a newer paper that describes the implementation of memory management units: Virtual Memory in Contemporary Microprocessors. It makes a good companion to our class paper.