TMR: Towards a Trusted MapReduce Infrastructure

MapReduce systems deployed over an open infrastructure such as a cloud have attracted much attention, due to the significant reductions in the costs entailed in satisfying both the computation and storage demands. However, in these systems, the integrity of MapReduce applications is subject to significant threats. Recent research mainly focuses on replication-based integrity verification schemes. However, inevitable critical deficiencies restrict its usage. In this paper, we propose a Trusted MapReduce (TMR) framework to integrate MapReduce systems with the TCG Trusted Computing infrastructure. TMR effectively uses remote attestations to achieve efficient and deterministic integrity verification. We propose a split and parallel attestation schema to reduce latency and eliminate scalability limitations when employing the Trusted Computing mechanisms. We implemented TMR on the Hadoop MapReduce system. Experiments showed that a high strength integrity assurance has been achieved, and the overheads can easily be managed to less than 1% for an industry-strength implementation.