MIME-Version: 1.0 Server: CERN/3.0 Date: Monday, 06-Jan-97 21:45:24 GMT Content-Type: text/html Content-Length: 8021 Last-Modified: Thursday, 11-Apr-96 15:24:04 GMT CS 380D : Distributed Computing I

CS 380D : Distributed Computing I

Spring 1996


Instructor : Lorenzo Alvisi

Teaching Assistant : Rajeev Joshi



Contents

Instructional Staff

Lorenzo Alvisi, Taylor Hall 4.122, Phone: 471-9792
Office Hours: Tuesdays, 10:00-12:00

Rajeev Joshi, UA-9 #4.108D , Phone: 471-9756
Office Hours: Mondays and Thursdays, 2:00-4:00 pm.

Other meetings with Lorenzo and Rajeev can be arranged by appointment.


Mechanics

I expect that 2/3 of the classes will cover material from the required textbook; the remainder will come from other sources (i.e. papers, other textbooks). References to such sources will be given in class at the appropriate time.

Lectures: 9:00-10:30 Monday and Wednesday, in Robert Lee Moore Hall 5.124.
The newsgroup for the class is utexas.class.cs380d.


Required Textbook

Distributed Systems, Second Edition, S. Mullender (editor), ACM Press, Addison-Wesley Publishing Company, Reading MA, 1994.


Course Content

CS380 covers abstractions that have proved useful or are expected to be useful for designing and building tomorrow's distributed systems. These include:

We will integrate the discussion of the general principles with the presentation of case studies that exemplify how such principles have been used to design and implement real systems. Other topics, depending on time and interest, will be presented by me or by some of you (the size of the class does not allow all of you to give a presentation). Such topics may include:


Grading

There will be 4 or 5 written homework assignment. Solutions will be graded F, B, or A. Any solution that demonstrates a credible effort on behalf of its authors (whether the solution is right or wrong) will receive a B or better.

Collaboration on homework assignment by up to three students is permitted and encouraged, but not required. When there is such a collaboration, a single solution should be submitted for grading, with the names of the collaborators. Other collaborations will be considered violations of Academic Integrity.

There will be a written, take-home midterm examination, for which no collaboration will be allowed.

There will be no final exam. Each student however will be required to write a final paper (about 20 pages) that surveys one of the issues that we have discussed in class. A list of suggested topics will be distributed in class on Monday 4/8. The paper is due at the start of the last class, Wednesday 5/1: hence, you will have 4 weeks to complete the paper.

You can also team up with a colleague and prepare one or two lectures on a topic not previously covered in class. If you choose this option, you and your colleague will only be required to write a single survey paper of about 20 pages. I warmly encourage you to consider volunteering for a presentation: it will give you an excellent opportunity to improve your communication skills.


Problem Sets

In this and all subsequent problem sets, you should conform to the following general guidelines:


Due: Mon, 5 Feb 1996
Problem 1
The snapshot protocols discussed in class and in the textbook assume that communication channels are FIFO. Derive a snapshot protocol for an asynchronous system that does not depend on the FIFO assumption, and prove it correct (i.e. prove that the protocol produces a consistent global state). You may assume that at most one snapshot is being computed at any point during a run.
Note: The book contains a reference to a paper by Mattern that contains a solution to the problem. I urge you to resist the temptation to solve the problem by visiting the library...

Problem 2
Taking the snapshot of a distributed computation is a general technique for computing stable global predicates. More efficient protocols can be derived for computing specific predicates, that are often conceptually simpler and more efficient (in terms of the number of messages they exchange) than a snapshot-based solution.

In this problem you are required to derive such a ``specialized'' protocol for detecting a deadlock in an asynchronous distributed system. Ideally, your protocol would not need a centralized monitor process, and would have a message cost of O(n), where n is the number of processes in the distributed system (a monitor-based snapshot protocol for detecting deadlock has a cost of O(n*n)).

The suggested solutions to these problems are now online. This link points to the postscript file.
Due: Wed, 28 Feb 1996, 0900
This link points to the postscript file describing the second homework assignment.

The final exam

The assignment constituting the final exam is due by 5 p.m., Friday May 3, 1996.
This link points to the Postscript file describing the assignment.
If you have questions, feel free to send email to Lorenzo or to Rajeev .

If you have ideas on improving this page, please send your suggestions to
joshi@cs.utexas.edu

Rajeev Joshi, last updated 11 Apr 1996