HyCC

Abstract

While secure multi-party computation (MPC) is a vibrant research topic and a multitude of practical MPC applications have been presented recently, their development is still a tedious task that requires expert knowledge. Previous works have made first steps in compiling high-level descriptions from various source descriptions into MPC protocols, but only looked at a limited set of protocols. In this work we present HyCC, a tool-chain for automated compilation of ANSI C programs into hybrid protocols that efficiently and securely combine multiple MPC protocols with optimizing compilation, scheduling, and partitioning. As a result, our compiled protocols are able to achieve performance numbers that are comparable to hand-built solutions. For the MiniONN neural network (Liu et al., CCS 2017), our compiler improves performance of the resulting protocol by more than a factor of 3. Thus, for the first time, highly efficient hybrid MPC becomes accessible for developers without cryptographic background.

Efficient MPC via Program Analysis

“…Finally, to demonstrate the practicality of our solver, we provide a toolchain that takes high-level unannotated source-Java source code in our case-, translates it to MPC-source, and outputs optimal protocol assignments. We compare our solver with publicly available benchmarks from [13,5]. We remark that the concrete assignment from [13,5] is known for only a subset of the related benchmarks; for all those we confirm the same (2-out-of-3 protocol) assignment.…”

“…However, since the optimal twoprotocol combination can be found in polynomial time, we can use our solver to compute best two-out-of-c protocols combinations for a constant c, by applying it to all possibles pairs of the c protocols, and picking the pair (and the corresponding solution) that minimizes the objective function across all c 2 applications. To demonstrate the quality of our OPA solver, we apply it to compute protocol assignments for known benchmarks from ABY [13] and HyCC [5], for which code has been released namely Modular Exponentiation, Biometric Matching and Private Set Intersection, Convolutional Neural Networks (CNNs) of MiniONN [23] and Crytonets [15], k-means clustering algorithm, DB Merge and DBJoin as well as new ones we introduce, namely Greatest Common Divisor (GCD) and Histogram. We remark that works [13,5] directly compute optimal assignment for three protocols (this is done by manual assignment in [13], and exhaustive search in [5]).…”

“…To demonstrate the quality of our OPA solver, we apply it to compute protocol assignments for known benchmarks from ABY [13] and HyCC [5], for which code has been released namely Modular Exponentiation, Biometric Matching and Private Set Intersection, Convolutional Neural Networks (CNNs) of MiniONN [23] and Crytonets [15], k-means clustering algorithm, DB Merge and DBJoin as well as new ones we introduce, namely Greatest Common Divisor (GCD) and Histogram. We remark that works [13,5] directly compute optimal assignment for three protocols (this is done by manual assignment in [13], and exhaustive search in [5]). However, with the exception of Modular Exponentiation in the LAN setting, all resulting optimal assignments use one protocol or a mixing of only two protocols.…”

“…Due to the difficulty of solving Integer Programming in general, this lead to a conjecture that the problem of optimal protocol assignment is NP-hard. The conjecture was adopted by follow-up works [26,8,5] and gave rise to heuristic approaches.…”

Recent Advances in Practical Secure Multi-Party Computation

“…Therefore, only a few specialists can implement MPC now, and this can be a barrier to wide usage in the real world. To break this situation, there are some results on MPC compilers [48], [80], [81]. MPC compilers take a (standard) program code (= function) written by major programming languages (e.g., C, Python) or domain specific language as input, and output a program code for MPC of the same function.…”