Reliable Probabilistic Gossip over Large-Scale Random Topologies

TL;DR

This paper introduces a new probabilistic gossip algorithm that improves reliability and reduces message complexity across various large-scale random network topologies, supported by a unified reliability modeling approach.

Contribution

A novel simple gossip algorithm with higher reliability and lower message complexity, along with a unified model to predict reliability across different random graph topologies.

Findings

The new algorithm outperforms existing gossip algorithms in reliability and message efficiency.

The reliability model accurately predicts the trade-off between reliability and message complexity.

Simulations confirm the model's effectiveness across diverse network topologies.

Abstract

This paper studies reliability of probabilistic neighbor-aware gossip algorithms over three well- known large-scale random topologies, namely Bernoulli (or Erd\H{o}s-R\'enyi) graph, the random geometric graph, and the scale-free graph. We propose a new and simple algorithm which ensures higher reliability at lower message complexity than the three families of gossip algorithms over every topology in our study. We also present a uniform approach to model the reliability of probabilistic gossip algorithms in the different random graphs, whose properties, in fact, are quite different. In our model a forwarding probability is derived with consideration of parameters in gossip algorithms and graph properties. Our simulations show that our model gives a reasonable prediction of the trade-off between reliability and message complexity for all probabilistic neighbor-aware gossip algorithms in various random networks. Therefore, it allows to fine-tune the input parameters in the gossip protocols to achieve a desirable reliability with tolerable message complexity.