Practical View-Change-Less Protocol through Rapid View Synchronization

The emergence of blockchain technology has renewed the interest in consensus-based resilient data management systems that can provide resilience to failures and can manage data between fully-independent parties. To maximize the performance of these systems, we have recently seen the development of several prototype consensus solutions that optimize for throughput at the expense of overall implementation complexity, latency, recovery costs, and the ability to deal with unreliable communication. Due to this, it remains unclear how these prototypes will perform under real-world workloads where latency and recovery matter. In this paper, we present the Practical View-Change-Less Protocol (PVP), a high-throughput consensus protocol that provides a stable and low latency in all cases, has a simple recovery path with low costs and can deal with unreliable communication. PVP does so by combining a chained consensus design, which can replicate requests with a reduced message complexity and that uses a novel Rapid View Synchronization protocol to enable robust and low-cost failure recovery, with a high-performance concurrent consensus architecture in which independent instances of the chained consensus can operate concurrently to process requests with high throughput and without single-replica bottlenecks. Due to the concurrent consensus architecture, PVP greatly outperforms traditional primary-backup consensus protocols such as PBFT by up to 430%, Narwhal by up to 296% and HotStuff by up to 1808%. PVP is even able to outperform RCC, a state-of-the-art high-throughput concurrent consensus protocol, by up to 23% in optimal conditions. Furthermore, PVP is able to maintain a stable and low latency and consistently high throughput even during failures.