Research And Implementation Of Dynamic Load Balancing And Transaction Management Model For Service Grid

Grids can integrate the distributed resources on the Web such as computing, data, storage, software and manpower to accomplish collaborative task. Grids started in the scientific computing but are now being deployed by business enterprises. Service grid is the new concept combining Web service and grid, and it will be a commercial focus in network economy pattern. Many businesses will be completely transformed over the next decade by using grid-enabled Web services to integrate across the Internet to share not only applications but also computer power. Under the support of the 863/CIMS projects we will apply the service grid to CIMS fields and establish a service grid prototype supporting enterprise flexible composition.In the business process scheduling of the service grid, many businesses will be integrated by using the grid service across Internet. So it needs to coordinate the distributed resources and carries through the job scheduling automatically and transparently by the local brokers. We should balance the load in service grid in order to enhance the resource's utility and increase system's throughput. We put forward the hierarchical dynamic load balancing schedule scheme and the load balance model consists of job dispatcher, scheduler, watcher and communication module. How to dispatch the job is the core mechanism in a load balance system. We design and implement a dynamic dual-threshold job dispatch algorithm, which takes into account the sums of the running-job, the performance and loading information of the local scheduler. It holds the characteristics of intelligent disposal, dynamic control, and load balancing.Service grid business is unreliable because of its character of loosely coupled, distributed and long running. The ability of providing reliability is the key that decides whether grid computing may be applied to commercial applications. However, missing so far from current technologies is the support of a facility to provide consistency and reliability for grid service applications. Transaction is a fundamental concept in building reliable distributed applications, and it insures all participants in an application to achieve a mutually agreed outcome. Referencing the transactionmanagement methods in Web service and the safepoint concept in workflow system, combining with the feature of the service grid, this paper presents a transaction management model, which is composed of coordination framework and coordination protocol. It can deal with short atomic transaction as well as long business transaction. We will introduce the concept of safepoint in workflow. The safepoint offers the possibility and flexibility to roll back only parts of a process instead of the entire process. This model can solve the problems of consistency and reliability efficiently in the processing of service grid applications.