High Speed Consensus with Trusted Execution Environments

Abstract

In recent years, Byzantine consensus algorithms have seen a surge in popularitywith the rise of Bitcoin and blockchain technology. A major problem that hampersadoption of existing consensus algorithms in blockchain scenarios is their scalability. There has been much research in the past years aiming to optimize thesealgorithms and increase their efficiency. For example, recent work has shown that voting rounds present in many classical algorithms can be made drastically more efficient by the use of message aggregation techniques.

Another trend is towards the usage of trusted hardware to increase performance and lower resource requirementsof these algorithms. Trusted hardware enables algorithms to reduce the lower bound on the number of replicas from $3f+1$ to $2f+1$, where $f$ is the number of tolerated faults. Currently, all existing Byzantine consensus algorithms either use no trusted hardware at all, or assume that all replicas have access to the same trusted hardware. This leaves a gap in the design space, neglecting scenarios where only some machines have access to trusted hardware.

In this work, we investigate the possibilities where only a subset of all replicas has access to trusted hardware. We introduce the SACBFT framework, consisting of two transformations that can be applied to existing Byzantine consensus protocols, increasing their efficiency by allowing them to make use of trusted hardware that exists in the system.

We apply the framework to PBFT and RePBFT to produce SACPBFT and SACRePBFT respectively, and show how to apply the framework to other protocols. We also evaluate a proof-of-concept implementation of SACPBFT, showing that it can dramatically reduce network usage and increase performance even when only a single replica has access to trusted hardware.