Abstract
This paper presents Cataphract, a batch processing method specialized for BFT databases. Batch processing is a common technique for byzantine fault-tolerant state machine replication (BFT SMR) and distributed databases. However, no batch processing method is optimized for BFT databases, which possess characteristics of both BFT SMR and distributed databases. Cataphract optimizes cryptographic and communication processing, which are bottlenecks in BFT databases. We evaluate Cataphract with Basil (state-of-the-art BFT database) in experiments. In an environment where nodes are within an availability zone, Basil with Cataphract demonstrates up to approximately 5.6 times higher throughput and reduces latency by up to about 55% compared to the vanilla Basil. In an environment where nodes are within a region, Basil with Cataphract demonstrates up to approximately 13.8 times higher throughput and reduces latency by up to about 74% compared to the vanilla Basil. In an environment where nodes are geographically distributed, Basil with Cataphract demonstrates up to approximately 80.4 times higher throughput and reduces latency by up to about 76% compared to the vanilla Basil.
