DEthna: Accurate Ethereum Network Topology Discovery with Marked Transactions

Chonghe Zhao, Yipeng Zhou, Shengli Zhang, Taotao Wang, Quan Z. Sheng, Song Guo

Introduction

Understanding the peer-to-peer (P2P) topology of the Ethereum network is critical for optimizing block propagation and improving scalability. However, Ethereum is designed to conceal its structure to mitigate targeted Denial-of-Service (DoS) and eclipse attacks. Existing discovery methods often suffer from low precision or high network overhead. To address these challenges, researchers introduced DEthna (Distributed Ethereum Network Analyzer) at INFOCOM 2024. DEthna utilizes a "marked transaction" strategy to accurately and efficiently map the Ethereum network by exploiting transaction replacement and propagation mechanisms.

The core of DEthna is its Link Inference Model, which determines the existence of a link between two target nodes, $A$ and $B$. This model relies on "marked transactions" and two fundamental Ethereum features: sequential nonce processing and transaction replacement (EIP-1559).

The "Marked Transaction" Strategy

DEthna issues four specific transactions ($TX_M, TX_A, TX_B, TX_C$) from a single account. The process involves two primary steps:

  1. Step 1 (Setup): Transactions $TX_A, TX_B,$ and $TX_C$ are sent to nodes $A, B,$ and the rest of the network ($C$) with a future nonce ($n+1$). Since the transaction with nonce $n$ is missing, these nodes "park" the transactions in their local queues without forwarding them.
  2. Step 2 (Trigger): DEthna sends $TX_M$ with nonce $n$. This "unlocks" the queue, prompting nodes to forward the $n+1$ transactions.

By carefully tuning gas prices, DEthna ensures that if the observer receives $TX_B$ from node $A$, it must have come directly from node $B$. This isolates the link and provides high-precision inference.

Single Link Inference Step and Process

Distributed Architecture

To scale topology discovery globally while minimizing costs and bypassing regional traffic blocking, DEthna employs a three-layer distributed architecture.

  • Controller Layer: A high-performance local server that manages the overall logic and coordinates the discovery process.
  • Relay Layer: Cloud-based nodes that assist in transaction distribution and data collection.
  • Performer Layer: A set of low-cost, lightweight cloud nodes distributed across various geographical regions (e.g., California, London, Singapore). These nodes probe different ISPs to ensure comprehensive network coverage.
DEthna's Three-Layer Distributed Architecture

Experimental Results and Insights

DEthna was evaluated on the Goerli test network, using a standard full node to establish ground truth. The results demonstrate significant improvements over existing baselines:

  • Precision: 0.983 (compared to 0.026 for K-bucket methods).
  • Recall: 0.889.
  • F1-Score: 0.934.

Beyond performance metrics, DEthna revealed critical insights into the Ethereum network's structure. Notably, over 50% of nodes exhibit low connectivity (16 or fewer connections), often due to being behind NATs or firewalls. This "long tail" of connectivity suggests that while the network is robust against random failures, it relies heavily on a small number of "super nodes," making it potentially vulnerable to targeted attacks.

Conclusion

DEthna provides a high-accuracy, low-overhead solution for mapping the Ethereum P2P topology. By leveraging valid transactions and the nonce-queue mechanism, it avoids the pitfalls of previous "spammy" discovery tools. Future work will likely involve adapting these methods as Ethereum's propagation protocols evolve to further obscure network structure.

Summary updated

BibTeX
@inproceedings{zhao2024dethna,
  author = {Chonghe Zhao and Yipeng Zhou and Shengli Zhang and Taotao Wang and Quan Z. Sheng and Song Guo},
  title = {{DEthna: Accurate Ethereum Network Topology Discovery with Marked Transactions}},
  booktitle = {IEEE INFOCOM - IEEE Conference on Computer Communications},
  year = {2024},
  url = {https://openreview.net/forum?id=7BBViHwEuC},
}
Figure