Emergent Peer-to-Peer Multi-Hub Topology

TL;DR

The paper tackles the challenge of building scalable, robust P2P overlays by enabling emergent hub structures without centralized control. It introduces Elevator, a gossip-based hub sampling protocol that blends preferential attachment (to promote hubs) with random attachment (to preserve resilience), yielding exactly $h$ hubs while each node maintains $c$ connections ($h$ hub links and $c-h$ random links). The authors define an API (init, getPeer, getHub), provide a detailed protocol with per-cycle and background operations, and validate performance through extensive simulations showing ultra-low diameter (≈$2$) and strong resilience to crashes, churn, and hub-targeted attacks. The results indicate that Elevator achieves stable hub-mediated connectivity with robust propagation properties, making it applicable to federated learning and decentralized validator selection while reducing reliance on static hub nodes. This work opens a pathway to hub sampling algorithms that balance efficiency and security in unstructured overlays, with practical implications for fast model dissemination and resilient decentralized systems.

Abstract

In this paper we propose and evaluate an innovative algorithm that enables the creation of Peer-to-Peer network overlays characterized by emergent multi-hubs. This approach generates overlays that balance between the randomness of a graph and the structure of a star network, resulting in networks that not only feature prominent hubs but also exhibit strong resilience to failures. By leveraging principles of preferential attachment and random attachment, our method allows hubs to form spontaneously, offering a decentralized and fault-tolerant solution ideal for applications requiring both low network diameter and high robustness. The protocol is entirely decentralized, operates asynchronously, and depends exclusively on local information. Nodes organically evolve into hubs and remain indistinguishable from other nodes (except in terms of the number of incoming links). The quantity of hubs that emerge can be predetermined by the application as a network parameter.

Figures (52)

• Figure 1: Clustering coefficient computed during the simulation (no failures), for each algorithm, every 10 cycles
• Figure 2: Average path length computed during the simulation (no failures), for each algorithm, every 10 cycles
• Figure 3: Diameter computed during the simulation (no failures), for each algorithm, every 10 cycles
• Figure 4: Clustering coefficient computed with a 50% crash, for each algorithm, every 10 cycles
• Figure 5: Average path length computed with a 50% crash, for each algorithm, every 10 cycles