A Study Of Multi-Source Data Distribution Mechanism For Publish/Subscribe System

The publish/subscribe system, which benefits from its loosely coupled messaging paradigm, is regarded as the key technology for the large scale high extensible network communication applications. In recent years, the applications of publish/subscribe system have been expanded into multi-source massive-amount data distribution fields. Existing data distribution mechanisms are mainly designed for single-source environments. Therefore, they are hardly suitable for the new features of the multi-source environment.To address this proplbem, a multi-source massive-amount data distribution mechanism is studied for the publish/subscribe system. The main work and contributions of the thesis include:(1)The process of multi-source massive-amount data distribution is divided into three stages, namely start-up, steady and ending stage. Four design principles are proposed to enhance the performance of data distribution mechanism in each stage. The first principle is to divide topic data into blocks with equal size and build the distribution tree for each block. The second principle is to ensure the interior-node-disjoint among all distribution trees. The third principle is to avoid nodes becoming the ancestors of nodes with lower load. The fourth principle is to balance the load by handing part of data forward tasks of overloaded nodes over to nodes with lower load.(2)Three typical data distribution mechanisms are migrated into publish/subscribe system as benchmarks and their pros and cons are well analysed.(3)A multi-source massive-amount data distribution mechanism called MeshCast is proposed. It conforms to the design principles mentioned above and is customized for the publish/subscribe system. Experiments among benchmarks and MeshCast are conducted.MeshCast utilizes the traits of multi-source massive-amount data distribution environment to enhance the data distribution efficiency. The theoretical model and the functional components of this new mechanism are elaborated. The strategies of load balancing and improving bandwidth utilization are also introduced. Contrast experiments manifest that MeshCast succesfully narrow the time consuming of multi-source massive-amount data distribution tasks.