| With the popularization of cloud computing, more and more concerns about the privacy and security issues are arising. Existing approaches on encrypted domain processing and trusted computing have been shown useful in enabling privacy-preserving computation, but have been found limited, impractical and expensive. Instead, this report focuses on another approach of anonymous routing between distributed untrusted devices. We analyze the potential channels of sensitive information leakage and propose a solution that achieves a strong guarantee of anonymity. We study how to execute distributed applications written in the popular MapReduce framework and Spark on an untrusted cloud. We design and implement our solution to add a mix phase/stage to enhance the existing Hadoop MapReduce and Spark framework. The evaluation shows it is feasible to run Spark jobs in our design with no change to the application logic. The performance cost of this privacy-preserving execution vary from different types of jobs. |
