Along with the development of grid application, the resources become fruitful rich, the performance and the reliability of network connection become unstable. But the convenience of grid environment also brings some security problems that don't exist in the network. Correctness of computing has become one of the hot topics of large-scale distributed computing systems in grid environment. Security and correctness of computing have become the bottleneck of the development of grid computing since malicious hosts and the third entities attack. Different users and different tasks need the different requirements to the correctness of computing and there are large amount of free computing resources in the grid environment. This is the paper's premise and background.In this paper, we analyze the advantages and the disadvantages of the existing model, for example redundancy checked model and replicated-agent migrant computation model with voting. We introduce a new concept of the task trust degree and show a new fault-tolerant model by means of trust degree mechanism and replicated-agent computation with voting. A user submits its task and its task trust degree to the grid environment according to its task's request. The job manager and the resource manager search hosts to satisfy the user's request and distribute the task to these hosts. Then it will compute the sum of the corresponding trust degrees with the same computing result, and make comparison and then select the result with the highest sum of trust degrees as correct one. In this paper, we make a simulated test that can verify the new fault-tolerant model. By the test data, we can draw a conclusion that the model have already reach the request in fault-tolerant.The new model makes full use of trust degree mechanism and voted method. It not only increases fault-tolerant capacity but also gives its users more choices for different tasks at computing cost. The hosts with higher trust degrees may get much benefit in the new model and computing resources may be sufficiently utilized. So we can say that the fault-tolerant model can realize the task fault-tolerant in some points and the capability that users can communicate with visual organization. |