Research On Group Communication For Cooperative Work In WANs

CSCW system supports the cooperative work of geographically distributed peoples. Like other distributed systems, messages will be transmitted to all active users in the reliable and ordered manner. Because it is not supported by the current network protocols, the study of group communication is introduced to manage the message transmission and the group organization. In past, the study was focused in the LANs, and recently extended to the WANs. Because there are still some shortages in system architectures and mechanisms, a novel system CSCSGROUP is proposed. At first, the system architecture is analyzed and designed. Because the previous systems leverage the idea that WANs are interconnected LANs, their framework designs are related to the LAN distributions. The applicability of the system is limited. Therefore, the client/server cluster architecture is introduced into our system. By grouping servers in the cluster independent of user distributions, the system can provide the services for any users with good scalability and wide applicability. Based on the new architecture, a set of transmission management mechanisms is designed according to virtually synchronous semantics. A novel fault tolerant algorithm is presented to eliminate the delay introduced into the message ordering process by the old method. In the management of group organization, unlike predefining algorithm setting the failure detection TIMEOUT value in advance, the forecasting algorithm is proposed to make it adaptive to the variable transmission states of WANs. Both the accurateness and the sensitivity can be obtained in failure detection. Based on it, the multi-cycle failure estimation algorithm is designed to avoid making wrong failure decisions in unstable member states and network states. In addition, the double event channels mechanism is used to tranfer member state change information timely. The awareness of the collaborators is enhanced. For balancing the load of server cluster, the classification algorithm is presented to describe the server workload more accurately. A modified Probabilistic Neural Network model is built in terms of the characteristic of server's service. Based on it the load balance strategy is designed. Finally, all above methods are realized as the function components. The adaptor interface is provided to construct the cooperative applications on CSCWGROUP system seamlessly. Unlike previous group communication systems inheriting the LAN algorithms, our system is designed directly to the WANs and can adapt well to unstable network state. Good multi-user communication environment is provided for cooperative work.