An Empirical Study of the I2P Anonymity Network and its Censorship Resistance

TL;DR

This paper empirically analyzes the I2P anonymity network, revealing its size, structure, and vulnerability to censorship, and discusses potential improvements for resistance against blocking techniques.

Contribution

It provides the first detailed empirical measurement of I2P's network properties and demonstrates its susceptibility to address-based censorship.

Findings

Approximately 32,000 active I2P peers daily.

A censor can block over 95% of peers with only 10 routers.

Blocking more than 70% of peers significantly impairs network usability.

Abstract

Tor and I2P are well-known anonymity networks used by many individuals to protect their online privacy and anonymity. Tor's centralized directory services facilitate the understanding of the Tor network, as well as the measurement and visualization of its structure through the Tor Metrics project. In contrast, I2P does not rely on centralized directory servers, and thus obtaining a complete view of the network is challenging. In this work, we conduct an empirical study of the I2P network, in which we measure properties including population, churn rate, router type, and the geographic distribution of I2P peers. We find that there are currently around 32K active I2P peers in the network on a daily basis. Of these peers, 14K are located behind NAT or firewalls. Using the collected network data, we examine the blocking resistance of I2P against a censor that wants to prevent access to I2P using address-based blocking techniques. Despite the decentralized characteristics of I2P, we discover that a censor can block more than 95% of peer IP addresses known by a stable I2P client by operating only 10 routers in the network. This amounts to severe network impairment: a blocking rate of more than 70% is enough to cause significant latency in web browsing activities, while blocking more than 90% of peer IP addresses can make the network unusable. Finally, we discuss the security consequences of the network being blocked, and directions for potential approaches to make I2P more resistant to blocking.