Stretch-BFT: Workload-Adaptive and Stretchable Consensus Protocol for Permissioned Blockchain

Although the leader-based consensus protocols, such as PBFT, are widely used in permissioned blockchains, the leader node may become the bottleneck of the system when the number of blockchain nodes or the system's throughput demand increases. On the contrary, the leaderless protocols deal with this problem by running multiple leader-based protocol instances concurrently. However, most of such works ignore the fact that deploying more instances is not always the best choice on some situations. When the system's workload is low, deploying more instances not only wastes resources, but may also compromises the ability of sluggish tolerance of the system (increasing latency). In this study, we propose Stretch-BFT, which dynamically adjusts the number of instances according to the system workload. Stretch-BFT includes three sub-protocols: 1) BFT workload sensing, 2) adaptive instances reconfiguration, and 3) failed instances recovery. The experimental results show that Stretch-BFT can exhibit high throughput as the existing leaderless protocol when the workload is relatively heavy, and improve the sluggish tolerance ability when the workload is relatively light.