Dynamical grouping model for distributed real time causal ordering

Abstract

This paper proposes a hierarchical architecture of communication allowing scalability in the case of multimedia applications such as teleconferencing where there is a large number of participants. From an unspecified communication network, we propose to decompose the system into local groups. Each local group is defined as a finite set of processes, which computes by performing some capacity tests. The proposed architecture guarantees a dynamic reconfiguration between local group members. Over this architecture, we present a real time synchronization protocol which allows to solve the problems of delay jitter and local clocks drift, and ensures real-time causal delivery. The protocol computes also the global delivery time of messages for the group multicast. We show that the proposed architecture reduces the buffer size and the use of the bandwidth compared to a direct connection architecture (complete graph between senders and receivers). To illustrate our purpose, we have implemented a tele-teaching application over LAN Emulation.