Improve availability of fault-tolerant computing: Optimal multi-task allocation in MapReduce

MapReduce emerges as a popular programming model for data-intensive scalable computing. As one of core components, task schedule comes to be a very hot topic in recent studies. However, the computing on fault-tolerant applications has not been covered yet. In the paper, we at first point out the importance of fault-tolerant computing and propose a novel model to find out an optimal task allocation scheme, which allows us to obtain the optimal job availability. To analyze the properties of task allocation, we proof several theorems and evaluate them with analysis and experiments. Our experiments show that the reduction of job unavailability by our method is about one order of magnitude as compared to the systematical allocation.