A Scheduling Policy With Real-time Characteristic For Service Grid

Grid is applied to full-scale connecting of all kinds of resources, such as computing resource, storage resource, communication resource, software resource, information resource and knowledge resource. The study of service grid is aimed at intellectualized information processing. The emphasis is to eliminate information acnodes and knowledge acnodes, ulteriorly to achieve general sharing of information and knowledge resource. Service grid is a rising technology based on the existent Internet. The basal establishments of service grid are the broadband digital communication networks, and wide-area distributed heterogeneous service nodes. A new Grid systematic framework is brought forward, which is quite different from traditional Client/Server architecture. The service grid system is composed of several autonomic domains, between which a flexibly coupled peer-to-peer structure is introduced, so that the distributed system obtains favorable automatic scalability. In each domain, a domain server and an information server is set. The domain server is a pivot in service grid, which is in charge of inner domain services and service proxies management, monitoring the state of service instances. Furthermore it is the key of implementation of real-time and fault tolerance scheduling. In order to provide steady services that meet certain QoS (quality of service) requirement, a scheduling policy with characteristic of fault-tolerance and real-time is advanced. According to user's material service request, grid middleware can search and match an optimal service from the information server and schedule task to the corresponding service node. During the processing of service request, a notification subscribing mechanism is taken to monitor instantaneous state of service instance. Once a failure occurs, the fault-tolerance scheduling module will be invoked to transplant the service task and rebuilt the service instance. A mechanism of QoS guarantee based on SLA is used to insure that the ultimate result meet user's original requirement. The whole system is implemented with Java on the basis of GT3. Graphic user interfaces are designed with Java Applet and JSP. The results of systemic test indicate that the real-time scheduling is feasible and effectual; furthermore, the fault-tolerance policy has greatly improved the availability and reliability of service grid.