Not eXactly Byzantine: Efficient and Resilient TEE-Based State Machine Replication

We propose, implement, and evaluate NxBFT, a resilient and efficient State Machine Replication protocol using Trusted Execution Environments (TEEs). NxBFT focuses on a "Not eXactly Byzantine" (NxB) operating model as a middle ground between crash and Byzantine fault tolerance. NxBFT's consensus layer is asynchronous, graph-based, leaderless, and optimized for the NxB operating model, enabling load-balancing of requests between replicas and, in fault-free cases, two network round trips between decisions. We identify fundamental issues with crash recovery due the use of TEEs in asynchrony that only can be circumvented by relying on synchrony for liveness. We provide a throughput-latency trade-off analysis of NxBFT, Chained-Damysus (rotating leader), and MinBFT (static leader) for up to 40 replicas and network round trip latencies up to 150 ms. NxBFT achieves the highest throughput in all scenarios. When small latencies are required, MinBFT and Damysus are at an advantage with Damysus benefiting from the NxB model in terms of throughput for small deployments. In contrast to leader-based approaches, NxBFT's performance is almost not impacted when actual crash faults occur.