Group Communication Protocols for Computer Networks

Supported in part by NSF grant NCR-9706285

Principal Investigator:

Philip K. McKinley
Department of Computer Science
Michigan State University

Project Description. As ATM networks are increasingly used both to directly support telecommunications services as well as to provide underlying backbone support to Internet applications, there is a pressing need for flexible and extensible group communication protocols for ATM networks. Important issues that must be addressed in providing group services include construction and maintenance of multipoint connections, membership management, and in some cases, consensus on a group leader. This research explores novel solutions to all three problems that take advantage of the underlying link-state routing (LSR) algorithm used to support point-to-point communication. Specifically, by modeling various aspects of group communication as consensus problems under LSR, robust and efficient solutions are developed to support multiple types of multipoint connections, fault-tolerant leader/membership consensus, and even improvements to LSR itself

Our preliminary research on two of these problems has achieved very promising results. First, new class of flooding protocols has been proposed for use in ATM networks. These switch-aided flooding (SAF) protocols take advantage of underlying ATM hardware cell relay and cell duplication to improve (in terms of bandwidth consumption, workload at switches, and flooding times) the performance of flooding operations. Second, a generic multipoint connection (GMC) protocol has been designed and evaluated. The GMC protocol operates independently of the chosen MC topology algorithm, enabling the use of many algorithms that are not supported by current multicast protocols. Both the SAF and GMC protocols are modeled as consensus problems under LSR, and simulation studies demonstrate their advantage over other approaches under typical network conditions.

Our ongoing investigations use the consensus model under LSR to address a much broader class of group communication problems in ATM networks. In particular, a novel approach to group communication is under study, which extends the signaling capabilities of groups and allows them to participate in communication activities in the same manner as hosts and switches. As with the preliminary studies, the primary methods used to study the proposed protocols will be analysis and simulation. Besides providing a better understanding of the issues and problems that must be addressed in supporting group communication in ATM networks, this project will produce tangible results in the form of new protocols and prototype implementations, which will be made available for use by other researchers and developers.