CS 552 Fall 2005

Class Project Assignment


The purpose of the class project is to introduce the research process to students. A successful project will propose and and evaluate an idea and/or discover and quantify an unknown property of an existing system. Projects can be with 1-2 students. 3 Student projects are possible, but must be larger in scope than a 1-2 person project.


Students will submit a proposal and final project report. The instructor will also ask the students for an informal update during the semester.


Below are possible project ideas.


  1. Quantification of the Discrepancy of Classic Queuing Theory. Many works have shown hat network packet traffic follows self-similar behavior in contrast to exponential behavior. However, classic queuing networks are still the simplest, easiest to use models of computer networks available. The goal of this project is to quantify the error of simple queuing models when traffic follows self-similar behavior. A successful project would quantify the discrepancy of a variety of properties obtainable from classic queuing theory such as the average time spent in a queue, average queue length, and drop probabilities, as a function of the self-similarity of the traffic. This project would likely use a combination of theoretical analysis and trace-driven simulation.

  2. Self-similar behavior: Man or Machine? An open question is the true cause of self-similar behavior. Given that human-generated processes have often been shown to observe Poisson-like behavior, one hypothesis is that machine-to-machine protocols generate packet traffic that is more self-similar than the root human causes. This project would quantify the self-similarity of a variety of human processes (e.g. Clicks and web-page views, file opens and closes, etc) and the resulting lower-layer processes generated from these (e.g. HTTP GETs, and packet traffic).

  3. Network Tomography. In this project you will replicate a network tomography study using PlanetLab. The goal of this study is to find a large graph of the layer 2 and 3 components in the network. Also, the project would describe the

  4. Internet Characteristics of the Top N sites vs. General Network. Several studies have made statements about Internet characteristics such as packet loss, jitter, availability and routing stability by comparing random pairs of hosts on the Internet. However, other studies of sites have found much different characteristics. In this study, you would measure the same network characteristics between any pairs of hosts and also between the top N destination sites on the Internet; the goal is to prove or disprove the hypothesis that the network infrastructure “tree” to the top N sites has much better characteristics than the network in general.

  5. Anonymity vs. Bandwidth. Is there a fundamental tension between Anonymity and bandwidth? The goal of this project is to articulate the tradeoff between these. A good project would not only compare existing systems, but also propose and evaluate ideas to improve or replace them.

  6. Economic Models for File Sharing and Content Distribution. The current economics of content distribution and ownership do not work well. Are there models of sharing beyond those already available? How can content creators and distributors be fairly compensated while at the same time allowing for legitimate fair use (e.g. parody, time shifting, and critical review)? What is the role of both technology (e.g. The size and cost of distribution networks as well as encryption and watermarks) in such a scheme? In this project, you would first formalize the problem (preferable through mathematics, but also possibly in simulation) and then argue why your solution is superior to what is current available.

  7. Design-time Quantitative “ility” Metrics. In this project, you would propose a quantitative measure of an “ility”. These include, but are not limited to, managability, diagnosability, repairability, maintainability, confidentiality. Your metric must the quantifiable, i.e., given 2 systems, you can compare them with a numeric rating, as well as be applicable at design time. A successful metric will also have a good intuitive mapping to real systems. You must apply your metric to several real systems.

  8. Long term load-generators for Internet Service Studies. An issue with many studies of Internet Services is the short time scales of these studies. Typically, studies under controlled conditions only last hours. At the scale of days to weeks, real systems are measured. However, the ability of the experimenter to vary the conditions during the experiment are quite limited on real systems. In this study, you would propose how to build realistic load generators that lasted weeks to months. You would also build a prototype and evaluate it running for at least 1 week, and show that it could last 10's of weeks.