Research And Implementation Of Group Communication System In Fault Tolerance Service

With demands on software reliability and security, how to guarantee systems survival even in the case of fails of parts of systems, is now becoming an improtant role in the design of high-performance & high-reliability software systems. Distributed system technique is one of the effective means to solve the problem. Theoretically, in an asynchronous system, it is impossible to distinguish a slow process from a faulty process. It leads to a fact that it is also impossible to solve consensus problem with a process fail in asynchoronous systems. Hence, how to design effective consensus protocols and algorithms is a hot point in distributed system research area.Group communication is for communication among group members. The kernel of group coomunication system for fault tolerance is consensus protocols and algorithms. The thesis firstly explores the architecture of group communication systems, consisting of message ordering and membership management. In order to survive failures of processes crashes and lost messages, the thesis proposes a new consensus protocol and algorithm based on the logical token mechanism. With a failure detector satisfying P property, the protocol can guarantee the delivery of requests in a total order among all replicas. Furtherly, an improved protocol with a weaker failure detector satisfying S property, is proposed and discussed. The thesis also compares the protocol with the Chandra&Toueg algorithm. Based on the consensus protocol, it puts forward a membership management protocol handling replica's joining & leaving, and group's combination & partition. At last, the thesis theoretically analyzes request delay time and the probability of system reliability in CSMA/CD network.The thesis narrates the implementation of the forementioned protocols, and extends the implementation in secure aspects in order to face the forged messages, tampered messages and so on. Performance tests shows that, in LAN environment, delay due to group communication middleware is not greater than 20 ms, and delay due to the secure one is below 40 ms.