Abstract
The ubiquitous nature of computers is driving a massive increase in the amount of data generated by humans and machines. The shift to cloud technologies is a paradigm change that offers considerable financial and administrative gains in the ef- fort to analyze these data. However, governmental and business institutions wanting to tap into these gains are concerned with security issues. The cloud presents new vulnerabilities and is dominated by new kinds of applications, which calls for new security solutions. In the direction of analyzing massive amounts of data, tools like MapReduce, Apache Storm, Dryad and higher-level scripting languages like Pig Latin and DryadLINQ have significantly improved corresponding tasks for software devel- opers. The equally important aspect of securing computations performed by these tools and ensuring confidentiality of data has seen very little support emerge for programmers.
In this dissertation, we present solutions to a. secure computations being run in the cloud by leveraging BFT replication coupled with fault isolation and b. se- cure data from being leaked by computing directly on encrypted data. For securing computations (a.), we leverage a combination of variable-degree clustering, approx- imated and offline output comparison, smart deployment, and separation of duty to achieve a parameterized tradeoff between fault tolerance and overhead in prac- tice. We demonstrate the low overhead achieved with our solution when securing data-flow computations expressed in Apache Pig, and Hadoop. Our solution allows assured computation with less than 10 percent latency overhead as shown by our evaluation. For securing data (b.), we present novel data flow analyses and program
xi
transformations for Pig Latin and Apache Storm, that automatically enable the ex- ecution of corresponding scripts on encrypted data. We avoid fully homomorphic encryption because of its prohibitively high cost; instead, in some cases, we rely on a minimal set of operations performed by the client. We present the algorithms used for this translation, and empirically demonstrate the practical performance of our approach as well as improvements for programmers in terms of the effort required to preserve data confidentiality.
