Toward a General Science of Secure Protocol Networks: Measuring and Improving Privacy, Security, and Resilience in Planet-Scale Decentralized Systems

Abstract

This dissertation develops a measurement-first science of secure protocol networks. These are planet-scale, decentralized, sociotechnical systems that coordinate behavior through standardized message formats and economic incentives. Protocol networks such as Ethereum, BitTorrent, DNS, and SMTP, function as hidden critical infrastructure that powers numerous applications and communities, yet their distributed control, fragmented telemetry, and multi-layer ownership leave major observability gaps in their security posture. The dissertation's central claim is that these systems cannot have robust security and resilience guarantees without first measuring the structures and dynamics that shape their risk. Three research questions organize the work: how to build credible longitudinal multi-vantage telescopes for protocol networks, which systemic mechanisms shape privacy, security, and resilience across protocol families, and which interventions remain practical, composable, and evaluable under real constraints. To answer those questions, the dissertation develops a set of reusable empirical instruments and analysis methods. MagnetDB is a 300-week longitudinal telescope for the BitTorrent DHT that captures 28.6 million torrents and 950.6 million files. EtherBee is a multimodal Ethereum node observatory deployed across ten geographically distributed vantage points, recording 130.9 million attack events and 45.4 million P2P sessions. An email inbox sensor captured email provenance over 361 days across 150 services, and NinjaDoH provides a build--measure--evaluate loop for adversarial intervention testing in censorship-resistant DNS. These artifacts are paired with a methodological framework built around multi-vantage observation, paired deployments, cross-layer panels, and longitudinal collection, as well as the Dataset Value Taxonomy (DVT) for reasoning about dataset maturity and stewardship. Across the empirical chapters, four recurring mechanisms emerge. First, visibility functions as an attack surface: a paired Ethereum deployment shows that beacon nodes receive roughly three times more hostile traffic than matched controls. This is a significant risk, as 25.68 % of publicly active beacon nodes on the Ethereum mainnet expose auxiliary services beyond their P2P ports. Second, de facto centralization is observed in protocol networks that are supposed to be logically decentralized. In email, the top eight autonomous systems carry 89.35 % of observed email volume. Third, overlay networks inherit physical chokepoints: Ethereum's traffic-weighted peer focal points converge toward the North America--Europe corridor, leaving peripheral regions such as South America effectively tethered to those core routes. Fourth, temporal dynamics shape security outcomes. In a study of the distribution of pirated media content on BitTorrent, torrents that appear on the network before the official release date of a media title carry a fourteen-fold elevation in suspicious executables. The dissertation then evaluates both authenticated overlay connectivity hardening and out-of-band snapshot recovery for Ethereum node operators. Using a discrete-time network simulator with 16,000 paired runs per scenario grounded in real-world incidents and stressors, the combined defense reduces median recovery time from approximately 3.4 hours to 0.6 hours in resynchronization scenarios and lowers tail-event exceedance rates from 10% to 3.0% under significant disruptions to Internet routing. Taken together, the dissertation contributes a cross-protocol taxonomy that links node visibility, de facto centralization, physical chokepoints, and temporal tail risk. It also contributes methodological advances in telescope-building and dataset stewardship formalized through the DVT, long-horizon empirical instruments, a modeling toolkit for inference under partial observability, and practical interventions. The broader conclusion is that secure protocol networks require measuring what is otherwise hidden, identifying the mechanisms that recur across ecosystems, and evaluating interventions against the same empirical baselines used to diagnose the problem.