Back
2024INF015
Doctoral thesis

Consensus in blockchain : from gossip to synchronous Byzantine fault tolerance

  • 2024

PhD: Università della Svizzera italiana

Distributed systems
Gossip protocols
Consensus protocols
Byzantine fault-tolerance
Blockchains
English State machine replication (SMR) and consensus are fundamental concepts in distributed systems, providing consistency, reliability, and fault tolerance. This thesis examines these problems in the context of modern decentralized systems, particularly blockchains. Specifically, it explores the integration of gossip communication with consensus protocols and investigates synchronous Byzantine fault-tolerant (BFT) consensus protocols. We first explore the impact of gossip-based communication on consensus protocols, using the Paxos algorithm as a case study. We introduce Semantic Gossip, which optimizes gossip communication through semantic filtering and aggregation. Experimental results demonstrate that Semantic Gossip reduces message overhead and improves performance while maintaining reliability. Next, we evaluate how synchrony violations impact the correctness of synchronous BFT consensus protocols, both with and without Byzantine attacks. We outline an experimental approach to determining a synchrony bound that, with high probability, prevents correctness violations. Applying this approach to a new protocol, BoundBFT, we find that communication diversity and redundancy enable BoundBFT to tolerate synchrony violations without compromising correctness, resulting in lower synchrony bounds and improved performance. Finally, motivated by experimental data on message delays, we present a hybrid synchronous system model that distinguishes between small and large messages. Within this model, we develop AlterBFT, a BFT consensus protocol that relies on the timely delivery of small messages for agreement while requiring large messages to be eventually timely to ensure progress. Our evaluation shows that AlterBFT achieves significantly lower latency than state-of-the-art synchronous protocols and offers comparable performance and higher resilience than state-of-the-art partially synchronous protocols. This thesis advances the understanding of consensus in partially connected and synchronous environments, providing practical solutions to improve the performance and fault tolerance of distributed systems.
Collections
Language
  • English
Classification
Computer science and technology
License
License undefined
Open access status
green
Identifiers
Persistent URL
https://n2t.net/ark:/12658/srd1330260
Statistics
Document views: 18 File downloads:
  • 2024INF015.pdf: 35