A Stochastic Programming Approach for an Enhanced Performance of a Multi-committees Byzantine Fault Tolerant Algorithm.

Byzantine fault-tolerance (BFT) algorithms enhance trustworthiness of distributed systems by guaranteeing their resilience to Byzantine faults. Traditional BFT algorithms suffer from scalability issues, resulting in performance bottlenecks (e.g., low throughputs) in large-scale distributed systems. Moreover, distributed systems are generally deployed on geographically and/or logically distributed networks, which aggravates the performance-scalability issue. To tackle this challenge, existing works have proposed a number of new BFT algorithms (e.g., HotStuff, FastBFT). However, limited work has explored parallel BFT based on a partitioned set of connected subgroups. This is challenging due to 1) heterogeneous communications delays between different, potentially geographically distributed, peers, and 2) peers may have a random crash and/or Byzantine failures, which contribute to the failure of the BFT consensus. To address these issues, we propose a stochastic programming (SP) model to maximise the throughput, while considering communications delays and failure behaviors as constraints. The SP model solution provides the optimal multi-committee organisation. Evaluation results show 24% throughput enhancement with the SP model.