Monday, March 25, 2002
11:00 AM
CSB 209
Jonathan Stanton
Johns Hopkins University
Practical Wide-Area Group Communication
Group communication systems have a successful history of providing useful services for the development of distributed applications. Several different semantic models for such systems have been proposed, analyzed, and studied, and a number of systems have been built and used in both commercial and research software systems. However, group communications has also been viewed by many as a niche service whose limitations make it useful for only very restricted classes of applications. Two of the most significant limitations have been their dependence on low-latency, high-bandwidth networks and the complexity or inflexibility of the systems.

This talk will discuss the challenges wide-area networks such as the Internet present for group communications and how these challenges were met by the design and implementation of a group messaging system called "Spread".

Specifically, I will discuss two aspects of the system. First, how overlay networks can effectively scale reliable group multicast, providing reliable wide-area message dissemination without requiring per-message end-to-end acknowledgments. Second, what unique problem arise when developing security protocols such as group key agreement and access control for fault-tolerant messaging systems and how we addressed these problems.

The talk will also present performance results for Spread on both local and wide-area networks.

The Spread system has been publicly released and used in various commercial and freely available applications for over 2 years. The types of applications range as widely as database replication, distributed event logging, online chat, distributed graphics rendering, and distributed certificate revocation management.