Research Of QoS Control Based On Resource Management In Grid

Grid is to integrate the resources in the whole networks into a large supercomputer. Gridcan fulfill the all-around sharing of such grid resources as CPU, disk, data, information,knowledge, experts, instruments, etc. It is a critical issue for grid to handle the ever-growingclient demands as well as to provide performance assurance and service differentiation fordifferent customers. Grid must have mechanisms and policies for establishing and supportingQuality of Service (QoS). Grid QoS is an important research field in grid and QoS is animportant factor for evaluating whether a grid application is successful or not.Presently, more and more attention has been paid to Grid QoS by the internationalresearchers and companies. Grid QoS is a very important research matter of many large gridprojects. However, the current Grid QoS research usually just focuses on only one aspect,handles a special question or guarantees only one performance value. That is, the present GridQoS can only provide single QoS mechanism for the customers. The grid environment isgeographically distributional, heterogeneous and dynamic, thus single QoS mechanism can’tsolve practical problems in grid environment.This dissertation thoroughly investigates the policies, models and performance analysis ofGrid QoS control. In addition, the prototype with QoS control mechanism is implemented. Thecontributions of this dissertation are presented as follows.A comprehensive model named QoS-aware resource management model is proposed. Inorder to provide an all-round QoS guarantee mechanism for the customers, this model combinesseveral QoS mechanisms, including customer job classification policies, resource descriptionmechanism, resource discovery polices, resource scheduling policies and job scheduling polices.These policies in this model are indispensable for implementing Grid QoS and they are tightlyassociated rather than isolated with each other.The advantages and disadvantages of two current resource discovery algorithms(centralized discovery algorithm, distributional discovery algorithm are analyzed. An advancedresource discovery algorithm based on routing transferring is proposed referring to the routingtransferring policies of network routers. The results of performance analysis show that thisalgorithm can adapt to the special characteristics of the grid. Through this algorithm, theresources needed by the customers can be discovered quickly. Futher more, this algorithm can reduce the response time of the system and can guarantee the QoS requirements of thecustomers.The resource scheduling and the job scheduling are respectively modeled by open queuingnetworks and closed queuing networks. The queuing networks models are evaluated with MeanValue Analysis effectively. Because the customers of grid have different priorities, the MeanValue Analysis used in the models is a modified analyzing algorithm named P-MVA. Thenumerical results show that the modified P-MVA can effectively analyze the systemperformance. The customers with higher priorities have better QoS than those with lowerpriorities.Considering that grid has the characteristic of simultaneous resource possession, thisdissertation proposed an approximate Mean Value Analysis algorithm (Simultaneous ResourcePossession-Mean Value Analysis with priorities, SRP-P-MVA) to evaluate the performance ofqueuing networks models of grid. The numerical results of performance analysis show thatSRP-P-MVA can correctly analyze the performance of the queuing networks models with thecharacteristic of simultaneous resource possession and can effectively guarantee the differentQoS requirements.