Design And Implementation Of Load Balancing Service In Multidatabase Of Massive Data

On the background of MDMP (Massive Data Management Platform), this thesis analyzes and compares the load balancing mechanisms provided in different layers, and discusses the middleware's load balancing service in detail. In allusion to the deficiencies of some middleware products of load balancing service, a type of load balancing service called AFLS (Application Fault-tolerated Load balancing Service) which follows the CORBA specification is designed and implemented.The primary contributions of this thesis include:1. We study policies and mechanism of load balancing, including interceptor technologies based on CORBA, the model of collecting loads, migration policies, and member selection strategies. Meanwhile, the model and the cost of the load balancing are also thoroughly discussed.2. we design a Framework which is CORBA-based extensible Application Fault-tolerated Load-balancing Service. Besides, the key technologies of designing the service are fully described.3. AFLS is implemented on the middleware platform of StarBus. Through adoption of double balancer structure, AFLS presents a solution to single point of failure existing in centralized load balancing service. Moreover, it implements clients request scheduling, load collecting and load migrating in a transparent way to application; it can provide dynamic management for multiple types redundant server, support application-defined load metrics and balancing policies, having good adaptability and extensibility.4. Finally, a group of experiments are done to test the function and efficiency of AFLS. Empirical results demonstrate that AFLS works correctly and the overhead of it is very small. All the requirements to AFLS are fully fulfilled. AFLS have been applied successfully in the certain national project.