SEPIA: Security through Private Information Aggregation

This paper investigates the practical usefulness of secure multiparty computation (MPC) techniques in multi-domain network management. We design and implement privacy-preserving protocols for event correlation and aggregation of network traffic statistics, such as addition of volume metrics, computation of feature entropy, and distinct item count. To improve the performance of our protocols, we design comparison operations that are optimized for large numbers of parallel invocations. The implementation of the protocols and the basic operations are made available in a library called SEPIA. We evaluate the running times and bandwidth requirements of our protocols with actual backbone traffic traces, both on a department-wide cluster and on PlanetLab. Our results show that the proposed protocols allow processing traffic in near real-time for up to 140 participants, depending on the protocol. Compared to implementations using existing general-purpose MPC frameworks, our protocols are significantly faster. In particular, event correlation which takes 3 minutes with SEPIA, would require around 2 days using existing frameworks.