The big data era has dramatically transformed our lives; however, security incidents such as data breaches put sensitive data (e.g. photos, identities, genomes) at risk. To protect users’ data privacy, there is a growing trend to build secure cloud computing systems, which enables computation over two or more parties’ sensitive data, while revealing nothing more than the results to the participating parties. Conceptually, privacy-preserving computing systems leverage cryptographic techniques (e.g. secure multiparty computation) and trusted hardware (e.g. secure processors) to instantiate a “secure” abstract machine consisting of a CPU and encrypted memory, so that an adversary cannot learn information through either the computation within the CPU or the data in the memory. Unfortunately, evidence has shown that, side channels (e.g. memory accesses, timing, and termination) in such a “secure” abstract machine may potentially leak highly sensitive information including cryptographic keys that form the root of trust for the secure systems.

This thesis broadly expands the investigation of a research direction called trace oblivious computation, where programming language techniques are employed to prevent side channel information leakage. To demonstrate the feasibility of trace oblivious computation, we have built several systems, including GhostRider, which is a hardware-software co-design to provide a hardware-based trace oblivious computing solution, SCVM, which is an automatic RAM-model secure computation system, and ObliVM, which is a programming framework to facilitate programmers to develop applications. All of these systems demonstrate a better performance than previous related system by one to several orders of magnitude.

Examining committee:

Chair: Dr. Michael Hicks

Co-chair: Dr. Elaine Shi

Dean’s rep: Dr. Lawrence Washington

Members: Dr. Zia Khan

         Dr. Charalampos Papamanthou