An Efficient Approach to Improve Security for MapReduce Computation in Cloud System

Running MapReduce computation in public cloud raises a series of security challenges since the service providers may not be properly protected. Due to the fact that the MapReduce applications are long-running, which increases the chance of an attacker to massively perform malicious attacks by exploiting the workers vulnerability, many workers may be compromised. Those workers could misbehave and thereby tamper the results integrity of all computations assigned to them. To tackle this challenge, this paper proposes an effective Result Verification Mechanism (RVM) using a reputation threshold-based voting method to ensure the result integrity of MapReduce on the map and reduce phases. Therefore, render the MapReduce computation accurate. Another major contribution of this paper is that we implement RVM based on Apache Hadoop and perform a series of experiments. The evaluation study of the experimental results demonstrate that RVM can significantly reduce computation overhead and guarantee a low error rate as compared to the simple voting method like m-first voting.