Small, stupid, and scalable: secure computing with faerieplay

How can Agnes trust a computation C occurring at Boris's computer? In particular, how can Agnes can trust that C is occurring without Boris even being able to observe its internal state? One way is for Agnes to house C in a strong tamper-protected secure coprocessor at Boris's site. However, this approach is not scalable: neither in terms of computation - once C gets larger than the coprocessor, it becomes vulnerable to Boris again - nor in terms of cost. In this paper, we report on our Faerieplay project: rather than worrying about the limited size of a secure coprocessor, we try to make it as small as possible, with limited RAM and CPU. We start with the Fairplay work of Malkhi et al on implementing Yao's blinded-circuit solution to secure multiparty computation with software - this permits Agnes to trust C, but is too inefficient for all but small C. We then use our own prior work on using trusted third parties for practical Private Information Retrieval to design and prototype tiny trusted third parties (TTTPs) that substantially reduce the overhead involved in blind circuit evaluation.