Poster: Cross Cloud MapReduce: an Uncheatable MapReduce

MapReduce [1] is becoming a popular data processing application on Cloud Environment. However, security issues make many customers reluctant to move their critical computation tasks to cloud. For instance, [2] points out a real security vulnerability that the cloud service leader Amazon EC2 suffers from: some members of EC2 can create and share Amazon Machine Image (AMI) to the EC2 community so that other users can deploy their server by simply loading an AMI. A malicious AMI, if widely used, could flood the community with hundreds of infected virtual instances. On the other hand, in MapReduce where jobs are carried out via the collaboration of a number of computing nodes, merely one malicious node may render the overall results useless. In a traditional MapReduce setting where each node is deployed on the cloud, the integrity of a computation can be easily compromised and difficult to detect. In this work, we propose a new MapReduce Framework, Cross Cloud MapReduce (CCMR), which can be deployed among a single private cloud and multiple public clouds. By using the replication, hold-and-test, verification and credit-based trust management approaches, CCMR can eliminate malicious compute nodes and guarantee high computation accuracy while incurring acceptable overhead.