If Iterative Diffusion Is The Answer, What Was The Question?

Abstract

A Distributed Concensus System (DCS) is a key component for a distributed system, maintaining and consenting over distributed state, as often required in distributed filesystems, cryptocurrencies and other applications. The prevalent mechanism to distribute the information needed to reach consensus in a DCS is through an iterative diffusion with the goal of distributing the (updated) information to at least the majority of DCS participants. Such diffusion is not just costly in maintaining the needed communication relations but also incurs latency in performing the diffusion itself. When looking closer at the goal of achieving the majority rule, however, one may wonder of the lack of multicast being used for achieving the same goal. This paper thus returns to the original intention of how to distribute information to a large set of users and outlines reasons for why multicast has not been widely used thus far. We evaluate the potential gain in doing so, based on an analytical model for the diffusion latency in an iterative diffusion DCS, bounded through empirical studies in an Ethereum system.