Efficitent client-transparent fault tolerance for video conferencing

As video conferencing plays an increasingly critical role in many business environments, there is a need to ensure highly reliable operation of the conferencing infrastructure. We present a scheme for adding fault tolerance to an existing video conferencing server. The scheme is client-transparent so that it can be used by the installed base of clients. While the scheme is based on replication, the associated overhead is negligible since the backup does not process the media streams. Most previous work on fault-tolerant network services focused on transaction-oriented services. Video conferencing is an interesting test-case for applying fault tolerance for other types of services since it combines critical conference state that must be protected with media streams where limited data losses are acceptable. Our implementation combines kernel modules with small changes to the server application to efficiently preserve both the reliable connections used for control messages and unreliable connections used for media transfer.