Reducing Replica Coordination in BFT Consensus through Dependable and Secure In-Network Message Ordering

Abstract

Programmable networks enable the deployment of in-network ordering services that can improve the performance of consensus protocols. However, existing approaches either tolerate only crash faults or require additional replica coordination to handle Byzantine failures, limiting their performance gains. In this work, we present NsoBFT, a Byzantine fault-tolerant (BFT) consensus protocol that leverages secure in-network ordering through a USIG (Unique Sequential Identifier Generator)-based message ordering service deployed at the network layer. The ordering service is implemented using Network Functions Virtualization (NFV), allowing it to operate as a virtualized network function that provides isolation and flexible integration into modern infrastructures. By combining NFV-based in-network sequencing with a USIG-protected sequencer, NsoBFT eliminates the extra replica confirmation step required by prior BFT in-network protocols, shortening the critical execution path while preserving consensus safety and liveness properties. We implemented NsoBFT and evaluated it against NeoBFT and NOPaxos, two representative in-network consensus protocols. Experimental results show that NsoBFT consistently achieves lower latency and higher throughput than NeoBFT, significantly narrowing the performance gap between crash-tolerant and Byzantine-tolerant in-network consensus.