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

2025-01-19

AI Summary

To address the challenge of balancing reliability and performance in state-machine replication under partial component failures, this paper proposes NxBFT: a leaderless asynchronous consensus protocol designed for a “partially Byzantine” fault model and built upon trusted execution environments (TEEs). Methodologically, NxBFT introduces a TEE-Rider architecture to achieve inherent load balancing and leverages TEE-embedded hardware randomness to realize an asynchronous common coin—ensuring liveness under low-load conditions. It further integrates crash-recovery support and request fingerprint-based deduplication. Experimental evaluation on a 40-node deployment demonstrates a throughput of 400 Kops/sec and end-to-end latency of approximately 1 second. Under failure scenarios, NxBFT significantly outperforms MinBFT and Damysus, validating the holistic advantages of its asynchronous TEE-coordinated design in terms of fault tolerance, efficiency, and robustness.

Technology Category

• Machine Learning: Hardware-aware ML
• Multiagent Systems: Agreement, Argumentation & Negotiation
• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust

Application Category

• Security and Privacy: Large-scale security measurements
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Abstract

We propose, implement, and evaluate NxBFT, a practical State Machine Replication protocol that tolerates minority corruptions by using Trusted Execution Environments (TEEs). NxBFT focuses on a ``Not eXactly Byzantine'' operating model as a middle ground between crash and Byzantine fault tolerance. NxBFT is designed as an asynchronous protocol except for liveness of setup and recovery. As a leaderless protocol based on TEE-Rider, it provides build-in load balancing in the number of replicas, which is in contrast to leader-based and leader-rotating approaches. With quadratic communication complexity, a TEE-based common coin as source of randomness, a crash recovery procedure, solutions for request deduplication, and progress in low-load scenarios, NxBFT achieves a throughput of 400 kOp/s at an average end-to-end-latency of 1 s for 40 replicas and shows competitive performance under faults. We provide a comparison with a leader-based (MinBFT) and a leader-rotating protocol (Damysus) and analyze benefits and challenges that result from the combination of asynchrony and TEEs.

Problem

• State Machine Replication
• Fault Tolerance
• Efficiency Optimization

Innovation

• NxBFT
• Trusted Execution Environments (TEEs)
• Asynchronous Operation