A replication-based mechanism for fault tolerance in MapReduce framework (Q1665600)

Summary: MapReduce is a programming model and an associated implementation for processing and generating large data sets with a parallel, distributed algorithm on a cluster. In cloud environment, node and task failure are no longer accidental but a common feature of large-scale systems. Current rescheduling-based fault tolerance method in MapReduce framework failed to fully consider the location of distributed data and the computation and storage overhead of rescheduling failure tasks. Thus, a single node failure will increase the completion time dramatically. In this paper, a replication-based mechanism is proposed, which takes both task and node failure into consideration. Experimental results show that, compared with default mechanism in Hadoop, our mechanism can significantly improve the performance at failure time, with more than 30\% decreasing in execution time.