Fault Tolerant Broadcast in Bandwidth-Constrained Networks

Abstract

This thesis addresses the problem of achieving scalable fault-tolerant broadcast in networks with limited bandwidth. We begin by examining the limitations of leaderbased protocols, such as HotStuff, which suffer from a leader bottleneck and reduced system throughput as the number of servers increases. To mitigate this, we propose CodedBcaster and Coded HotStuff, a Byzantine Fault Tolerant (BFT) broadcast scheme based on erasure coding, demonstrating a significant improvement in throughput. We further explore the problem of optimal rate allocation in heterogeneous node-constrained networks and provide concrete theoretical results for determining the optimal system throughput rate. Additionally, we propose the MaxMin Rate Controller (MaxMin-RC) protocol as a feedback-based solution to optimize broadcast throughput in non-BFT settings, achieving close alignment with the optimal throughput rate. Through extensive simulations and evaluations, we demonstrate the effectiveness of our proposed solutions.