Research Of Adaptive Fault-tolerant Strategy Based On Dynamic Scheduling In Grid

Grid computing is a new branch of distributing computing, whose objective is to integrate the resources that are geographically dispersed and maybe belong to different organizations. What the grid computing could provide is a high performance, high transparence and high reliability unified computing environment. Since the characteristics of highly distributive, heterogeneous and dynamic of resources, it has been a difficult problem to guarantee the reliability of grid computing.Most of the present research in fault-tolerant grid mainly aimed at fault detecting and not much research has been taken in fault handlings. Moreover, the common fault detecting techniques were typical to some special applications. The traditional fault-tolerant techniques were faced with great challenges as varying of the complexity and dynamic in grid applications.Aiming at the requirement of fault-tolerance, the paper analyzed the present fault-tolerant grid systems and compared their advantages and disadvantages, then proposed an Adaptive Fault-Tolerant Grid System Model based on these researches. The paper gave a description of the architecture, design idea and basic frame of the model in detail, especially the main modules of the model, like information service, scheduling service, fault detecting service and fault managing service. Then a further discussion was taken in several key techniques. This paper put forward an adaptive fault detecting algorithm, which can export the probability of fault at any time. It allows users to set the rank of fault detection for different level fault tolerances, which can satisfy the requirements for specific applications. Then, the paper applied a strategy based on QoS on the choice of fault handlings. By formalizing several fault handlings, the paper analyzed their advantages and disadvantages basing on the performance of executing time, expenses and reliability, then utilized a simple method to select the most suitable fault handling according to the users'requests and the fault handlings'authority. In the last part of the paper, experiments were taken to validate the efficiency of the adaptive algorithm. The algorithm was proved to be feasible and effective by analyzing the experiments'results and comparing with other algorithms.