PAReS: A Proactive and Adaptive Redundant System for MapReduce

Abstract

Recently, MapReduce has been a key and popular technology for tackling data-intensive applications. But its two master servers in current MapReduce implementations have a single-point-of-failure problem, which may interrupt MapReduce operations and filesystem services. In this paper, we propose a hybrid takeover scheme, called PAReS (Proactive and Adaptive Redundant System). To further improve the master servers\' service qualities, PAReS also employs three service-quality improvement mechanisms, including a proactive synchronization and replication method, a mutual monitoring algorithm, and an adaptive warm-up mechanism. The extensive experiments show that PAReS enhances service quality at acceptable energy consumption level and synchronization cost as compared with four state-of-the-art schemes.