Privacy and Reliability in an Untrusted Cloud

Опубликовано 27 июля 2016, 19:31

The cloud offers unprecedented access to computation. However, if the underlying cloud may be malicious, ensuring the privacy and reliability of that computation remains a significant challenge. How do you keep the data private even from the cloud itself? In this talk, I will describe two ideas, a crazy one on achieving privacy through computation decomposition, and a less-crazy one on achieving reliability via smart redundancy. Together, these ideas help rein in the cloud, even when faced with Byzantine faults and malicious attacks. Both ideas challenge the accepted software development and distributed systems norms to produce exciting benefits (albeit often at the cost of computational efficiency). I'll describe prototype implementations of each system and empirical evidence of performance and reliability, as well as theoretical evidence of privacy.