Projects

Title	Researchers
Anonymity and Performance in Peer-to-Peer Systems	Nikita Borisov and Jason Waddle
Persistence of Data in a Dynamic, Unreliable Network	Rachel Rubin and Hakim Weatherspoon
Introduction of Anonymous Data to the Gnutella System	Benjamin Poon and Jonathan Hess
An atomic data middleware for managed networks	David Ratajczak and Sonesh Surana
Distributed XML Filtering Using a Structured Peer-to-Peer Overlay	Shariq Rizvi and Paul Burstein
DHT-Based Distributed Crawler	Owen Cooper, Sailesh Krishnamurthy, and Boon Thau Loo
Replica Placement Game	Marco Barreno and Byung-Gon Chun
Overlay Architecture for End-host Control against Flooding Attacks	Karthik Lakshminarayanan and Jayanthkumar Kannan
Implementing Eddy Tuple Routing in the PIER Query Routing System	Ryan Huebsch and Shawn Jeffery

Still looking for a project? Here are some project suggestions.

PIER-related (contact Joe Hellerstein, jmhATcs.berkeley.edu)

Mining/Analysis of Gnutella queries.
The Gnutella study we have done so far focuses on recall/search overheads. It is interesting to do some textual analysis on the actual queries. There have been studies on click streams from web logs, but not anything that I know of of mining p2p file sharing query streams. Some interesting questions to find out may be how many queries would benefit from fuzzy searches, how many typos, repeat queries, whether queries differ by region etc, etc. Also, figure out good online ranking functions for file sharing query results.
Net monitoring under stress.
PIER is a query processor based on DHTs. We are proposing to use PIER for p2p Internet health monitoring. This task is most important when the network is under stress -- during worms and other attacks. In these scenarios, we'd like the DHT to degrade gracefully, and the query engine to perhaps back off on its quality of service in various ways (e.g. via query approximation or rewriting) to avoid overloading the network further.
Near-neighbor queries for net-health monitoring in PIER.
As we deploy PIER for p2p Internet Health monitoring, a natural question users will have will be "how does my local performance compare to people like me?" This is a vague query, and it can be paramaterized in a number of ways. The general solution for any parameterization is to run "near-neighbor" queries to fetch readings from nodes "like me" and summarize them. Both the specifics of the appropriate queries and the implementation of generic near-neighbor facilities raise challenges.
P2P network trace mining.
The MINDS project at Minnesota (http://www.cs.umn.edu/research/minds/MINDS.htm) has shown how to use simple data mining techniques (anomaly detection and association) to aid in network administration. How might those techniques be made to work in a p2p environment?