Load balancing and fault tolerance are two key concerns when developing distributed computing applications. The emergence of grid computing environments extends this problem. Based on distributed scheduling architecture, a grid service oriented scheduling model is proposed to enable load balancing in this paper. In this model, the functionalities are characterized by fault tolerance, and implemented by applying intelligent agent technology. Degree of credit is introduced to make that a decision of grid service selection, service instance selection/migration and service instance redundancy, context agents are responsible for sensing the context of grid, and balance agents are enforced to automatically serve the scheduler for dynamic load adjustment and service-level fault tolerance. Between agents, information exchange is conducted via an agent communication language.