Research On The Resource Scheduling Technologies For Scientific Data Grid

Features of the grid include the large number of nodes, the equal relation among nodes and the autonomy of each node. It is facing more and more difficulties to achieve an effective resources scheduling in such a loose coupling distributed system. This dissertation researches on the resource scheduling technologies of the Scientific Data Grid and has received supports from the project of"Scientific Data Grid", a project of the national high technology development 863 program of China, and the project of"Scientific Database and Its Application System", a project of the CAS'information plan.The contributions of the dissertation as follows:1) A distributed resource scheduling model that adapt to the data grid. This dissertation has developed a distributed resource scheduling model. In this model, the same global resource scheduling software has installed on each node. By the software, each node can receive tasks submitted by grid users. Each node act as both a resource node and a grid portal. Every two nodes can communicate to each other. Thus, a node can carry negotiation with other nodes by exchanging messages and a suite of related nodes can make a coordinated scheduling.2) A Virtual resource Pool based (VP-based) grid resource scheduling mechanism. Since there are generally many nodes in a grid system, a single grid application cannot make use of all nodes for itself. When received a task, the resource scheduling system firstly choose a node set which has a lower cost and the then carries out the resource selection and scheduling within this set. This set is defined to be a Virtual resource Pool. In the fully distributed resource scheduling architecture, this dissertation propounds a VP-based grid resource scheduling mechanism. Its kernel are scheduling with a VP and the P2P principle within a VP.3) Four VP algorithms. This dissertation has developed four VP algorithms: VP initial node algorithm, VP set-up algorithm, VP task/node mapping algorithm and VP failure recovery algorithm.4) A method for getting the optimal number of computing nodes of a VP. This dissertation has an in-depth study on the optimal number of computing nodes of a VP and has gotten formulas on variable premises. By the formulas, we concludes that members of computing nodes of a VP is decided by the grid application of this VP, computing nodes that have already been selected and the candidate computing node.5) SimSDG and related experiments. In order to evaluate the resource scheduling...