VBFT: Veloce Byzantine Fault Tolerant Consensus for Blockchains
Mohammad M. Jalalzai, Chen Feng, Victoria Lemieux
TL;DR
VBFT introduces a two-step Byzantine fault-tolerant consensus for blockchains under partial synchrony. It attains optimal resilience with $n=3f+1$ and $f=\frac{n-1}{3}$, requires no trusted hardware, uses signature aggregation, and relies on QC, TC, and NC certificates to enable efficient view changes with $O(n)$ authenticator verification. VBFT guarantees strong safety for honest leaders and supports rotating leaders; a Byzantine leader's proposed block can be revoked during view changes if committed by fewer than $f+1$ honest nodes, but safety against double-spending is preserved and misbehavior leads to blacklisting. Liveness is ensured through exponential back-off, timeout propagation, and probabilistic leader selection, enabling progress after a finite number of view changes. VBFT thus offers low-latency consensus with full resilience and efficient recovery, making it well-suited for WAN blockchain deployments.
Abstract
Low latency is one of the most desirable features of partially synchronous Byzantine consensus protocols. Existing low-latency protocols have achieved consensus with just two communication steps by reducing the maximum number of faults the protocol can tolerate (from $f = \frac{n-1}{3}$ to $f = \frac{n+1}{5}$), \textcolor{black}{by relaxing protocol safety guarantees}, or by using trusted hardware like Trusted Execution Environment. Furthermore, these two-step protocols don't support rotating leaders and low-cost view change (leader replacement), which are important features of many blockchain use cases. In this paper, we propose a protocol called VBFT which achieves consensus in just two communication steps without sacrificing desirable features. In particular, VBFT tolerates $f = \frac{n-1}{3}$ faults (which is the best possible), guarantees strong safety for honest leaders, and requires no trusted hardware. Moreover, VBFT supports leader rotation and low-cost view change, thereby improving prior art on multiple axes.