Performance Comparison of Two Generic MPC-frameworks with Symmetric Ciphers

Thomas Lorünser, Florian Wohner


Research on multiparty computation (MPC) made substantial progress over recent years. It can be used to protect the privacy of data and users in modern application scenarios like Blockchain and the Internet of Things where different stakeholders want to collaborate. In this work we analyze practical aspects of two generic MPC frameworks, MP-SPDZ and MPyC, to generate new insights into the state-of-the art for generic and platform independent MPC. We implemented various symmetric ciphers and did extensive benchmarking on both frameworks to see how universal and generic they are and if they can be used without special knowledge. We found that the achieved performance cannot be trivially estimated from the algorithms without implementing. The stream cipher Trivium was by far the fastest and most portable in our tests. Contrary to most of existing literature we also addressed non optimal network settings and found surprising results. The asynchronous architecture of MPyC turned out to make more efficient use of the network layer in scenarios with higher network latency and loss and could even compensate for the optimizing compiler used by MP-SPDZ.


Paper Citation