Parallel and Cloud-Native Secure Multi-Party Computation

Abstract

Secure multi-party computation (MPC) enables analysis based on sensitive data from multiple data owners, applying distributed cryptographic protocols to ensure privacy. Such protocols introduce distinct communication requirements, causing the computation to run significantly longer than its counterpart, conventional computing. General MPC frameworks are available that make it simple to develop such privacy-preserving applications, but running said applications assumes multiple non-colluding computing parties that host the protocol runtimes, having rigorously set up the required infrastructure. Utilising cloud resources for this occasion is a good alternative to on-premises deployments. First, it allows for a larger degree of automation in the infrastructure set-up. Secondly, cloud datacenters enjoy superior network characteristics, detrimental for MPC performance, and offer elastic compute resources at competitive price models. This thesis presents a cloud-native deployment of the SHAREMIND MPC framework on Kubernetes. It further proposes methods for parallel programming, with which MPC applications could be scaled over clusters. Familiar programming models, MapReduce and bulk-synchronous parallel, are adapted to MPC, and benchmarked in commodity clouds, showing near-linear speedup.