Consensus crash testing: Exploring ripple's decentralization degree in adversarial environments

Abstract

The inception of Bitcoin as a peer-to-peer payment system, and its underlying blockchain data-structure and protocol, has led to an increased interest in deploying scalable and reliable distributed-ledger systems that build on robust consensus protocols. A critical requirement of such systems is to provide enough fault tolerance in the presence of adversarial attacks or network faults. This is essential to guarantee liveness when the network does not behave as expected and ensure that the underlying nodes agree on a unique order of transactions over a shared state. In comparison with traditional distributed systems, the deployment of a distributed-ledger system should take into account the hidden game theoretical aspects of such protocols, where actors are competing with each other in an environment which is likely to experience various well-motivated malicious and adversarial attacks. Firstly, this paper discusses the fundamental principles of existing consensus protocols in the context of both permissioned and permissionless distributed-ledger systems. The main contribution of this work deals with observations from experimenting with Ripple's consensus protocol as it is embodied in the XRP Ledger. The main experimental finding suggests that, when a low percentage of malicious nodes is present, the centralization degree of the network can be significantly relaxed ensuring low convergence times. Those findings are of particular importance when engineering a consensus algorithm that would like to balance security with decentralization.