Research On Techniques Of Scalable Computer Network Management System

As the high performance computer network systems with reliability and security become increasingly important in the recent years, the computer network management system attracts more attentions. Traditional centralized network management systems have shown inefficiency to manage the modern networks that are growing dramatically in size and complexity. Accordingly, the approaches that can improve the scalability of the network management system are required.In this dissertation, we make researches on several important technologies of the scalable network management system, which include system architecture, failure detection for network-based distributed applications(e.g., the distributed network management system), and network security management. The main contributions of this thesis can be summarized as follows:Techniques to improve the scalability of the CARD(Content-Aware Requires Distribution) method is studied, and a new architecture for building scalable network management system based on the CARD-DWSS ( CARD-based Distributed Web Server System) technology is proposed. The web-based distributed network management system used to extend its management capability by adding middle level servers and dividing the management responsibilities among them, while providing a centralized user interface from the top level server. When the interaction information or the interaction frequency between the user and the web-based network management system grows tremendously, the top level server will become the system bottleneck. In the new architecture, all the management servers are organized through the content-aware requests distribution mechanism to provide the user interface together. As a result, all phases of the management operations, including the delivery and the execution of the management operation commands, the collecting, processing and displaying of the network management information, are distributed completely, and excellent scalability can be achieved.An advanced gossip-style failure detection method is put forward to detect the failure of the network-based distributed applications, Although the gossip-style failure detection method can provide better performance than others when the size of the detected system increases, it becomes slow when the proportion of the failure members becomes large. The new failure detection method suggested by this thesis addresses this problem by...