A multiprocessor scheduling algorithm for low overhead fault-tolerance

We propose a new scheduling algorithm for achieving fault tolerance in multiprocessor systems. The new algorithm partitions a parallel program into subsets of tasks based on some characteristics of a task graph. Then for each subset, the algorithm duplicates and schedules its tasks successively. Applying the proposed algorithm to three kinds of practical task graphs (Gaussian elimination, Laplace equation solver and LU decomposition), we conduct simulations. Experimental results show that fault tolerance can be achieved at the cost of a small degree of time redundancy, and that performance in the case of a processor failure is improved compared to a previous algorithm.