美联储-Heraclius_一个具有现代支付系统潜力的拜占庭容错数据库系统（英）-2025_15页_574kb

Summary of Heraclius: A Byzantine Fault Tolerant Database System

Overview

Heraclius is a scalable, Byzantine Fault Tolerant (BFT) key-value store designed for use in modern payment systems.
It leverages parallelism and cryptographic proofs to provide fault tolerance against malicious attacks and silent data corruption, enhancing security and resiliency compared to crash fault tolerant (CFT) systems.

Key Features

• Architecture: A two-layer system with coordinator and shard clusters, using leader-based BFT consensus (e.g., Smart-BFT library) to handle transactions in parallel.
  • Multiple clusters work together to increase throughput, coordinated through inter-cluster communication using Merkle trees for efficient verification.
  • Tolerates faults via a quorum requirement (2f+1 nodes for f faults), ensuring correctness with cryptographic attestations.
• Scalability: Achieves parallel processing by partitioning data and requests, allowing horizontal scaling.
  • Innovations include cryptographic proofs to validate cross-cluster requests without excessive message overhead.

Performance

• Prototype implementation (1-64 clusters, 4 nodes each):
  • Peak throughput: ~110,000 operations per second with 0.2-second latency.
  • Scalability plateaus at 32 clusters; adding more clusters does not improve throughput beyond this point due to overhead.

Comparison

• BFT vs. CFT Systems:
  • BFT systems (e.g., Heraclius) offer better security against Byzantine faults and no reliance on external mitigations, but at the cost of higher operational complexity and lower scalability than CFT systems (e.g., Project Hamilton, PArSEC).
  • CFT systems support higher throughput (up to ~1.4 million operations/second in prior work) but lack protection against malicious nodes or bugs.
• Prior Work:
  • Compared to systems like Bitcoin (low throughput, energy-intensive) and Hotstuff/BFT-SMaRT (single-cluster limitations), Heraclius enables parallel scaling via multiple clusters.
  • Achieves lower latency (0.2s) than Bitcoin but fewer operations/second than optimized CFT systems like PArSEC.

Tradeoffs

• Advantages: Enhanced fault tolerance, security, and suitability for high-risk payment systems.
• Disadvantages: Higher node requirements (more nodes than CFT for equivalent throughput), increased complexity in implementation and maintenance, and operational costs.
• Decision Factors: BFT is preferable where resiliency is paramount and threat models include malicious attacks; CFT is better for applications prioritizing scalability and cost-efficiency.

Conclusion

• Heraclius demonstrates that a scalable BFT database can be built but faces scalability limits, achieving sufficient throughput for some payments systems.
• The tradeoffs between BFT and CFT highlight that neither design is universally optimal; the choice depends on the specific requirements of the system (e.g., security vs. scalability).

Future Work

• Expand performance tests to include varied workloads, concurrency levels, and support for transactions spanning shards.
• Explore two-phase commit for cross-shard transactions and integrate smart contract capabilities at the coordinator layer.
• Investigate advancements in Zero Knowledge Proofs (ZKPs) to reduce overhead and improve scalability.