A collaborated framework of grid computing was well defined, every grid node would be able to share resources and collaborate with each other. In grid environment, every dynamically used resources should be monitored and analyzed. How should we monitor and allocate dynamic resources in grid is an important issue all the time.
In this thesis, there are two different types of process scheduling model, one is non-grouped grid node process scheduling model and the other one is grouped grid node process scheduling model. With non-grouped grid node process scheduling model, users could assign jobs at any node in the grid, and this node would dispatch those jobs based on collected information of resources and job scheduling algorithm. That is to say the overall performance of grid environment would be improved because non-grouped grid node process scheduling model could monitor and allocate dynamic resources. However, there is a CPU consumption caused by monitoring dynamic resources in grid. The non-grouped grid node process scheduling model is not suitable when there are more than fifty nodes in the grid environment. With grouped grid node process scheduling model, nodes would be classified into five classifications by the CPU specification scores. Within each classification, every node in the same classification would be assigned to a group which is no more than ten nodes in that group. By separating nodes into groups, the CPU consumption caused by monitoring the resources of nodes in a group would be effectively reduced. While the group-agents node deal with all the resource allocation between groups, the utility ratio of the grid resource would be increase. Grouped process scheduling grid node model could be implemented in the grid with more than fifty nodes. By implementing these two types of process scheduling model, the overall performance of grid would be improved.
