An End-to-End Privacy Preserving Design of a Map-Reduce Framework

We present a secure design for a map-reduce framework that guarantees preservation of privacy of the original data. We use Hadoop as a typical environment for illustration. That is, in spite of the data divisions/replications for the computations, the privacy of the original data remains invariant. Specifically, we use the novel information flow model called RWFM model that assures that in spite of data divisions/replications that get triggered in map-reduce computations, the original data providers' policies are fully preserved through a dynamic labelling of data. Thus, our secure framework can be adapted for computing on MLS data preserving its' confidentiality and privacy specifications. This is realized through an automatic dynamic labelling through the RWFM model. We describe a design and establish that it preserves the security and privacy of the original data and illustrate the approach through examples.