Optimizing MPC for Robust and Scalable Integer and Floating-Point Arithmetic

Secure multiparty computation (SMC) is a rapidly maturing field, but its number of practical applications so far has been small. Most existing applications have been run on small data volumes with the exception of a recent study processing tens of millions of education and tax records. For practical usability, SMC frameworks must be able to work with large collections of data and perform reliably under such conditions. In this work we demonstrate that with the help of our recently developed tools and some optimizations, the Sharemind secure computation framework is capable of executing tens of millions integer operations or hundreds of thousands floating-point operations per second. We also demonstrate robustness in handling a billion integer inputs and a million floating-point inputs in parallel. Such capabilities are absolutely necessary for real world deployments.