CRII: SaTC: Towards Cost-Efficient Private Computation for Cloud Data Science

About This Grant

Collaborative data science leads to scientific advancement: the better the data, the better the science. Cryptographic protocols such as secure multiparty computation (MPC) enable collaborative data science on private datasets for richer scientific discovery, especially in fields with downstream societal relevance such as genomics. MPC, however, incurs non-trivial computational overhead; so, in the face of increasing cloud computing costs, data scientists are disincentivized from adopting MPC. The objective of this project is to advance knowledge in private, collaborative computation by developing techniques that reduce the cost of practical MPC for data science in the cloud. The project's novelties are providing valuable insights into the real-world characteristics of private computation on data science tasks and instantiating the first cost-efficient protocol for cloud MPC. The project's broader significance and importance are minimizing the costs of generating the transformative scientific insights enabled by private, collaborative data science, thereby making judicious use of limited resources to achieve the greatest impact. The project builds MPC on top of spot instances, which are heavily discounted cloud instances with the caveat that they can be revoked (or shut down) at any time. The project studies how dynamic MPC -- a theoretical primitive that allows computing nodes to enter and exit during MPC -- can be deployed to handle this revocation. The project provides the first comprehensive analysis of dynamic MPC in practice, evaluating these protocols on representative data science workloads. Then, it applies these experimental insights to develop the first MPC system specifically tailored for spot instances, adapting dynamic MPC techniques to continue computation even when spot nodes are revoked. By building MPC on top of spot instances, the project lays the foundation for an open-source, cost-efficient platform for large-scale data science over sensitive datasets, which will empower greater scientific discovery without compromising privacy. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.