Fault Handling for Multi-party Real-time Communication

For real-time communication services to achieve widespread usage, it is important that the network services behave gracefully if any component(s) fail. While other researchers have previously considered failure-handling for non-real-time communication as well as for unicast real-time communication, these failure-recovery techniques must be reexamined in the light of the changes introduced by the new protocols and services for supporting multi-party real-time communication. In this report, we describe techniques and mechanisms for maintaining network services for multi-party real-time communication in the face of failures that may make parts of the network inaccessible. The key goal is that the protocols should provide high performance in the common case (i.e., in absence of failed components) and the network performance should gracefully degrade in face of network failures; e.g., in the presence of network faults, the routes selected may not be as good, the connection setup may take a little more time, or resource allocation may be less eecient. We describe appropriate policies for storing state in the network, as well as the mechanisms for re-establishing connectivity for previously established connections and to permit setting up new connections to existing conferences. We also describe a redundancy-based approach, using forward error correction (FEC), and dispersing the FEC'ed data among disjoint routes. With these mechanisms, we can make multi-party real-time communication protocols robust to single and/or multiple failures in the network, without diluting the strength of the performance guarantees ooered, or sacriing the system performance in the common case, i.e., when all components work correctly.