Research On Key Techniques Of Resource Reservation Based On Resource Availability Evaluation In Computational Grids

As the key technology for the global resources sharing and coordination, grid computing will be widely used in the future. In grid environment, the dynamic, heterogeneous, and autonomous features of resources challenges the stability of QoS(quality of service) provided for grid users. Advanced resources reservation (RR) mechanism is implied in real grid systems to provide QoS guarantee for end users. However, the uncertainty of grid service makes it difficult to realize ideal performance of RR. Furthermore, RR has many negative effects on the performance of gird system and the local task execution. Therefore, study on resource reservation mechanism is both theoretically sound and practically useful, which takes the profits of each participant into accounts, and controls its negative effects with the aim of providing "non-trivial QoS" effectivelyIn order to assure QoS of grid tasks, RR mechanism must guarantee reliability of the resources during the reserved period and degrade the negative effects of RR. This thesis focuses on the RR that can provide effective QoS guarantee of grid tasks, degrade negative effects on local task, and increases reliability of reservation admission control, on the basis of analyzing risk of reservation violation, participants'interest, resources price, and reliability of reservation admission. The main contributions and achievements of this thesis are as follows.1. The user-oriented balanced benefit-risk RR modelThe uncertainty of resource performance is the biggest obstacle to ensure SLA can be guaranteed and realize the goal of the RR. The benefit-risk model is established based on the analysis of reservation violation and reservation benefit. The model can evaluate the performance of resources node in RR through benefit and risk, and can predict the benefit and risk based on mathematical expectation and variance theory respectively. The double task programming in the model can be transformed into convex quadratic programming through utility function, and the reservation schema characterized by balanced benefit-risk (BBR) can be achieved. Through the adjustment of risk degree, the preference of resources users for interest and risk can be satisfied, and RR can adapt to the actual grid environment. Simulation experiment shows that, compared with the currently employed RR strategy, BBR has an obvious advantage in task failure rate, resources utilization rate, and some other performance index.2. The utility-driven RR strategyThe Utinity-driven reservation strategy (UDS) with multi-node coordination is presented based on the analysis of resources price, market competitiveness of resources, and beneficiary of RR. Through service quality analysis of undergoing resources performance, the service quality satisfaction degree can be achieved, and the functional relation between the quality of resources service and market competitiveness is established. Under market economy environment, income of resources owners can be adjusted by lever of price, and a coordination reservation schema that can balance the reservation benefit of various resources can be achieved. Theoretical analysis and simulation experiments indicate that, compared with traditional reservation strategy, UDS strategy is preferential in the resources nodes load, balancing the relative interest rates of resources nodes, and enhancing the QoS.3. Reservation mechanism aiming at balance between the interest of resources provider and security of users' QoSTraditional RR strategy mainly focuses on the interests of resources provider or resources user, or considers the QoS for the user as the sole goal. Through the interests analysis of RR related participants, the Cost-and-Profit-balancing reservation mechanism (CPRM) is presented to balance the security of grid users'QoS and interest of resources provider. Based on the statistical features of local tasks, the resources pricing methods under different local management strategies can be achieved. By transforming the non-cooperative game between local task and grid computing into the maximization problem of system interests, the RR algorithm is presented. The simulation experiment shows that compared with traditional RR strategy, CPRM can proportionate the resources node load and enhance the utility rate of free resources.4. More reliable and empowered RR strategy The dynamism of grid resources node fluctuate the ability of resources service, which makes the decision of RR acceptance very tough. RRS suitable for the preference of grid user is introduced based on the stability analysis of resources nodes RR acceptance. Resources service ability can be obtained through the analysis of local resources load and the actual performance of accepted RR task. The reliability of RR acceptance can be assessed by the probability of performing the RR tasks. The tolerance degree of grid user for RR risk may be satisfied by adjusting the threshold value of RR acceptance. Simulation experiment shows that RRS works well in the assurance of RR performance, esp. when the request rate of RR tasks increases from15%to25%, the expected breach rate can be5%lower than the traditional RR strategy.