Study On Fault-Tolerant Atomic Broadcast Based On Switched Ethernet

With the development of computer science and network technology, more and more study and research are focused on distributed system and parallel computing environment. One of the most important issue of distributed system and parallel computing environment is how to guarantee reliable broadcast in the cooperation between distributed processes, especially in distributed database system, distributed directory service and so on.Over the last decade, many reliable broadcast protocols have been proposed, including totally ordering broadcast protocol, fault-tolerant broadcast protocol and atomic broadcast protocol.As the importance of reliable broadcast, broadcast protocol for different kinds of networks have been raised such as a kind of reliable broadcast architecture for parallel cluster computing environment, reliable broadcasting in faulty hypercube computers, a reliable broadcast algorithm with selected acknowledgements in mobile ad hoc networks, and so on.At the same time, switched Ethernet technology are rapidly developed as the emergence of programmable switch, layer 3 and layer 4 switching. Switched Ethernet has many advantages compared to traditional share media Ethernet like its real-time property. And now switched Ethernet are widely used in process control, factory automation and other applications that support high bandwidth real-time environment.Switched Ethernet use IEEE802.1d distributed spanning tree discovery algorithm to avoid "broadcast storm" problem in layer 2 routing. Switches in switched Ethernet running spanning tree algorithm will organize a spanning tree top structure to guarantee a unique communication link between any two switches.In this paper, we propose a new fault-tolerant atomic broadcast algorithm based on switched Ethernet. This algorithm mainly uses the information of spanning tree top structure to maintain a logical token ring and token passing. And at the same time, this algorithm employs a batched broadcast and a batched acknowledgement to reduce the bandwidth needed for broadcast and acknowledgement by any broadcast process and provide fault-tolerant atomic broadcast to the upper applications.