TL;DR
This paper provides a comprehensive analysis of the network structure and resilience of seven major blockchain (BC) overlay networks. By crawling these networks and constructing detailed connectivity graphs, the authors reveal that even a small number of highly connected nodes can be targeted to partition major BCs, and that overlapping nodes participating in multiple BCs present a significant vulnerability to the entire ecosystem.
What Problem Does This Paper Address?
Blockchain systems are lauded for their decentralized nature and robustness against attacks. However, their actual resilience heavily depends on the underlying peer-to-peer (P2P) overlay networks that connect the nodes. Despite the widespread adoption of blockchains, the critical design aspects, connectivity properties, and network-layer interdependencies of these P2P overlays remain poorly understood. This lack of understanding makes it difficult to assess their security and resilience against various threats. This paper aims to fill this gap by systematically studying the structural properties and resilience of seven prominent BC networks to targeted attacks.
Methods
The authors developed a novel, efficient, and non-intrusive method to study BC overlay networks. Instead of relying on traditional, often infeasible, topology inference techniques that require high connectivity or transaction fees, they adopted a "probe and crawl" approach. Their crawler, modified from an existing tool, probes BC networks every two hours for 28 days, collecting information about peers and their known connections. This allows them to construct connectivity graphs representing potential connections between peers, rather than just established ones. They then analyze these graphs using standard network analysis metrics and evaluate network resilience by simulating targeted attacks, such as removing nodes based on their degree or betweenness centrality.
Datasets
The study utilized data collected from seven distinct blockchain networks: Bitcoin, Bitcoin Cash, Dash, Dogecoin, Ethereum, Litecoin, and Zcash. The data collection spanned approximately one month (June 26 - July 22, 2020), resulting in 335 network snapshots for each BC network, totaling 2345 graphs. This dataset represents a novel benchmark for analyzing BC overlay network structures and resilience.
Key Insights and Ideas
- Hidden Interconnections: A significant finding is the presence of "overlapping nodes" – peers that participate in multiple BC networks simultaneously. This overlap, which can range from 10-40% of nodes in a network, creates a critical vulnerability. Attacking these nodes can disrupt multiple BCs concurrently, posing a systemic risk.
- Targeted Attacks are Devastating: The research demonstrates that major BCs like Bitcoin can be partitioned into disconnected components by removing fewer than 10 highly-connected peers. This highlights that while BCs are robust against random failures, they are highly susceptible to well-informed, targeted attacks.
- BC Networks are Unique: Unlike traditional networks like the Internet or social networks, BC overlays exhibit distinct structural properties. They are not purely random and don't fit neatly into well-known network models, suggesting a need for specialized analysis.
- High Up-time Nodes are Key Targets: There's a strong positive correlation between a node's session length (how long it stays online) and its degree. This means nodes that are consistently online are also highly connected, making them prime targets for attackers seeking to disrupt network connectivity.
Key Results
- Network Structure: BC networks vary in density, with larger networks generally being less dense. Their degree distributions are heavy-tailed and belong to the exponential family, but they don't exhibit small-world properties despite low diameters.
- Resilience: While generally resilient to random node removals, targeted attacks can significantly impact BC networks. For instance, removing just 5 central nodes in Bitcoin or Bitcoin Cash can reduce the largest connected component by 60%.
- Overlapping Node Impact: Targeting overlapping nodes, while less effective than targeting top central nodes within a single BC, offers the advantage of disrupting multiple networks simultaneously. This strategy can significantly shrink the largest connected component in several BCs.
- Temporal Dynamics: BC networks are dynamic, with variations in degree distribution observed across snapshots. However, some networks like Dash and Dogecoin show more stable temporal characteristics.
Limitations and Future Work
The authors acknowledge that their method infers potential connections, and actual connections might be a subset. They also note that recent Bitcoin Core updates might affect the accuracy of future topology inference. Future work could explore more sophisticated attack models, investigate the impact of network churn on resilience, and develop strategies to enhance network robustness beyond simple topology hiding. The study also suggests the need for further research into the implications of overlapping nodes for the overall security and stability of the blockchain ecosystem.
Summary updated