Measurements, Analyses, and Insights on the Entire Ethereum Blockchain Network

Measurements, Analyses, and Insights on the Entire Ethereum Blockchain Network

A summary of the WebConf 2020 (formerly WWW) research paper by Xi Tong Lee, Arijit Khan, Sourav Sen Gupta, Yu Hann Ong, and Xuan Liu.

Background of the Ethereum Blockchain: It has been more than ten years since Bitcoin [1] introduced the era of decentralized community-controlled currency. Since then, several cryptocurrency variants like Litecoin, Namecoin, Dash, Zcash, have been introduced. Blockchains are increasingly becoming popular due to the prevalence of such cryptocurrencies and decentralized applications. Decentralized applications are written on the framework of numerous blockchain networks like Hyperledger, Corda, Ripple, Stellar, EOS, NEO, IOTA, and many more. Among them, Ethereum [2] is a distributed public blockchain network that focuses on running code (smart contracts) for decentralized applications. More simply, it is a platform for sharing information in a global state that cannot be manipulated or changed.

While Bitcoin-like cryptocurrency networks concern themselves only with users (wallets) transacting over blockchain, Ethereum-like blockchains present a decentralized computing environment. Ethereum is a transaction-based state machine, where the state is made up of accounts. Transfer of value and information between accounts cause transitions in the global state of Ethereum, which are recorded in the blockchain [3]. There are primarily two types of accounts: (a) User accounts, controlled by external users with their private keys, and (b) Contract accounts, controlled by contract codes that behave like “autonomous agents”. Transactions in Ethereum are data packets sent by the user accounts, signed with their private keys, while Messages in Ethereum are virtual objects produced by contract accounts, generally sent to other contracts.

In addition to the native unit of value ether, Ethereum blockchain allows creation of Tokens, an abstraction of “digital assets”, with the help of suitable data structures and methods implemented through smart contracts. Similar to base transactions using ether, accounts in Ethereum may transact in tokens of various kinds, fungible or otherwise, through the appropriate smart contracts. This allows for a rich ecosystem of tokens, including various ERC20 (fungible) and ERC721 (non-fungible) tokens, to thrive on Ethereum blockchain.

Motivation of Our Study: The genre of blockchain introduced by Ethereum brings forth a fascinating ecosystem of humans and autonomous agents (smart contracts), cohabiting the underlying blockchain fabric. It is neither like conventional social networks, where the players are human users, nor like the cryptocurrencies, where all interactions are transfer of value or asset. In essence, a blockchain network is closer to the Internet or Web, where users are allowed to interact with one another, as well as with programs. However, different from Web, there is also an interaction framework for smart contracts, where they can call (or kill) each other to maintain and advance the global state of the blockchain. This motivates us to study a public permissionless blockchain network as a complex system. We choose Ethereum, the most prominent public permissionless blockchain, to measure and draw insights from the network interactions.

In the last decade, several works [4-9] explored Bitcoin, cryptocurrencies, and other blockchain networks based on graph theory and network analysis. This line of research gained momentum due to the transparency offered by public permissionless blockchain, which allows anyone to access transactional information on the networks. We follow the direction of these previous works to measure and analyse the entire Ethereum blockchain network — above and beyond the financially relevant token transfer layer. Our approach closely follows the norms of measuring and analyzing social networks, Internet, and the Web [10-23], as the entirety of the Ethereum blockchain network presents itself as an equally complex system.

Datasets and Four Constructed Networks: Google Cloud BigQuery curates the entire Ethereum blockchain data in terms of blocks, contracts, transactions, traces, logs, tokens and token transfers [24]. We extract all relevant data for Ethereum from the ethereum_blockchain dataset under the Google Cloud bigquery-public-data repository, till 2019-02-07 00:00:27 UTC, which amounts to all blocks from genesis (#0) up to #7185508.

Figure 1: Interactions in the Ethereum Blockchain Network

We are interested in all interactions between Ethereum accounts, both in terms of standard ether transactions and token transfers. This requires us to construct interaction network from the Ethereum blockchain data, where vertices are accounts (users or contracts) and arcs denote their interactions. There are four major types of interaction between Ethereum addresses — (i) User-to-User (transaction or token transfer), (ii) User-to-Contract (call or kill), (iii) Contract-to-User (transaction or token transfer), and (iv) Contract-to-Contract (create, call, kill or hard fork), as illustrated in Figure 1. In addition, there are some interactions to and from the Null address, which denote creation of smart contracts and generation of ether (mining rewards), respectively. We create four interaction networks out of the entire blockchain dataset of Ethereum, as follows.

TraceNet. We first create TraceNet, with all possible user and smart contract addresses found in the entire blockchain dataset as vertices, and all successful traces with non-null from/to addresses as arcs. We characterize the vertices by their type – regular users, miners/ mining pools, regular contracts, and miner/ mining pool contracts. This is the most comprehensive interaction network for Ethereum, with various well-classified vertices and arcs.

ContractNet. The second network, ContractNet, is a subgraph of TraceNet, where we retain the arcs with both from_address and to_address belonging to smart contracts (verified using the contracts table). This provides us with a pure contract-to-contract interaction network on Ethereum, where arcs are direct messages and/or transactions between smart contracts. We observe four arc types in this category — (i) Create arcs that involves creation of new smart contracts, (ii) Suicide arcs where the owner of the smart contract decides to kill the smart contract, (iii) Call arcs that transfer ether from one account to another or call another smart contract, and (iv) Daofork arcs where a hard fork has occurred in the blockchain. These arcs connect three major vertex types — (i) ERC20 token contracts, (ii) ERC721 token contracts, and (iii) other contracts for intermediary functions and token-related services.

TransactionNet. The third network, TransactionNet, is the network of all Ethereum transactions recorded in the transactions table. Transactions are made by users, either to other users or smart contracts, or to a Null address in case of smart contract creation. The vertices and arcs of this network are thus similar to that of the TraceNet, with the exception of the Null address as an extra vertex in the network, and the ‘User to Null’ arcs.

TokenNet. Finally, we create TokenNet, pertaining only to the transfer of tokens between Ethereum accounts. We use the token transfers table to extract only token related transactions in the blockchain. The basic types of users and arcs are somewhat similar to that in the TransactionNet, with an additional level of arc characterization based on the token in use.

We have open sourced our network datasets here [25]. Each interaction network provides us with a different perspective on the Ethereum blockchain, and our analyses on the networks reveal new insights by combining information from the four networks. While TraceNet presents a global view of interactions between Ethereum accounts, ContractNet focuses only on the automated multi-agent network of contracts, providing us with a functional view of the Ethereum state machine. While TransactionNet helps us analyze the base ether transactions

in the blockchain, TokenNet focusses on the rich and diverse token ecosystem built on top of the Ethereum blockchain.

Summary of Contributions: To the best of our knowledge, we are the first to conduct a comprehensive study of the large-scale Ethereum blockchain network, cohabited by both human users and autonomous agents (smart contracts).

— We study the four blockchain networks based on local and global graph properties, e.g., network size, density, degree distribution, in-to-out degree correlation, vertex centrality, reciprocity, assortativity, connected components, core decomposition, transitivity, clustering coefficient, higher-order motifs, articulation points, adhesion, cohesion, and small-world characterization. Such structural information is useful to characterize interactions, to evaluate current Ethereum-blockchain-system at scale – an effort that has not been attempted before. We also identify their similarities and differences with social networks and the Web, and draw interesting conclusions.

—We further consider three prominent token subnetworks, Bancor, Binance Coin, and Zilliqa, and investigate the amount of activity in the token network over time, as well as the size of the core community driving the token economy over time. We identify interesting correlation between the temporal evolutions of the number of cores in the token subgraphs against the corresponding evolution of price of the token in the cryptocurrency market.

— We open source the datasets [25] and highlight important research directions such as analysis of mining pools, identifying complex patterns to detect fraudulent activities, and temporal analysis of token subnetworks to forecast the price of Ethereum backed tokens.

Our Findings and Future Research Directions: We investigate several local and global graph properties over four Ethereum blockchain networks (TraceNet, ContractNet, TransactionNet, and TokenNet), as well as in three prominent token subnetworks (Bancor, Binance Coin, and Zilliqa), and conduct a thorough experimental evaluation.

We find that these blockchain networks are very different from social networks. In case of both TraceNet and TransactionNet, Lognormal, Weibull, and Power-law with cut-off are better fit than the traditional power-law degree distribution. In all four blockchain networks, we have higher outdegree vertices (e.g., mining pools and mixers), as well as higher indegree vertices (e.g., ICO smart contracts). This characteristic is similar to the Web, consisting of both hub (having higher outdegrees) and authority (with higher indegrees) vertices, and is unlikely in social networks, which usually have high correlation between indegrees and outdegrees. As a result, blockchain networks are disassortative, having very low transitivity. Moreover, most frequent motifs observed in blockchain graphs are chain and star-shaped. Complex patterns, such as triangles, cycles, and cliques occur less, indicating lack of community structure in blockchain networks. Removal of only the highest-degree vertex (e.g., Binance, a global cryptocurrency exchange) can disconnect the entire largest weakly connected components in these graphs.

In spite of the aforementioned differences, blockchain networks are surprisingly small-world and well-connected. Analogous to social networks, blockchain graphs have average shortest path lengths only 4∼6. Similar to both social networks and the Web, blockchain networks contain a single, large strongly connected component (SCC), and about 98% of the remaining vertices can either reach this SCC, or can be reached from the SCC.

In terms of the four different networks, we observe that ContractNet has more self-loops and bidirectional arcs (hence, higher reciprocity), while TokenNet has fewer of them. As a result, the MultiDigraph of ContractNet is denser, while the simple, undirected version of TokenNet is more dense. Both of them yield larger core indices for vertices in the innermost cores, indicating higher density of their innermost cores. Moreover, both ContractNet and TokenNet have smaller radius and diameter compared to our larger networks, TraceNet and TransactionNet.

Following our characterization of Ethereum into four different blockchain networks, there is ample opportunity for future work. Study of individual mining pools as complex self-contained evolving networks would be interesting, as would be an investigation on the interplay between mining pools to identify instances of selfish mining and mining strategies. Further analysis of the individual Token networks in terms of activity signatures and temporal evolution (like change in coreness) may lead to more accurate forecasting of trading behavior and token prices in the cryptocurrency market. Identification of influential vertices and complex motifs (like cliques and cycles) in the blockchain networks may also lead to detection of fraudulent activities in the transaction and token networks of Ethereum. Quite naturally, a similar line of measurements and analyses can be applied to other public blockchain platforms to unearth interesting phenomena within and across the Web of blockchain networks.

For more information, click here for our paper.  Datasets: GitHub.

Blog post contributed by: Arijit Khan

  

[Reference]

[1] S. Nakamoto. 2008. Bitcoin: A Peer-to-Peer Electronic Cash System. https://bitcoin.org/bitcoin.pdf.

[2] F. Vogelsteller, V. Buterin, et al. [n.d.]. Ethereum Whitepaper. https://github.com/ethereum/wiki/wiki/White-Paper.

[3] G. Wood. [n.d.]. Ethereum: A Secure Decentralised Generalised Transaction Ledger. https://github.com/ethereum/yellowpaper.

[4] L. Ermann, K. M. Frahm, and D. L. Shepelyansky. 2018. Google Matrix of Bitcoin Network. The European Physical Journal B (2018), 91–127.

[5] S. Ferretti and G. D’Angelo. 2019. On the Ethereum Blockchain Structure: A Complex Networks Theory Perspective. Concurrency and Computation: Practice and Experience (2019), e5493.

[6] D. Ron and A. Shamir. 2013. Quantitative Analysis of the Full Bitcoin Transaction Graph. In Financial Cryptography and Data Security.

[7] S. Somin, G. Gordon, and Y. Altshuler. 2018. Network Analysis of ERC20 Tokens Trading on Ethereum Blockchain, In Complex Systems. Springer Proceedings in Complexity, 439–450.

[8] S. Somin, G. Gordon, and Y. Altshuler. 2018. Social Signals in the Ethereum Trading Network. CoRR abs/1805.12097 (2018). arXiv:1805.12097 http://arxiv.org/abs/1805.12097.

[9] F. Victor and B. K. Lüders. 2019. Measuring Ethereum-based ERC20 Token Networks. In Financial Cryptography and Data Security.

[10] L. Adamic, O. Buyukkokten, and E. Adar. 2003. A Social Network Caught in the Web. First Monday 8, 6 (2003).

[11] Y.-Y. Ahn, S. Han, H. Kwak, S. Moon, and H. Jeong. 2007. Analysis of Topological Characteristics of Huge Online Social Networking Services. In WWW.

[12] L. Backstrom, D. Huttenlocher, J. Kleinberg, and X. Lan. 2006. Group Formation in Large Social Networks: Membership, Growth, and Evolution. In KDD.

[13] A.-L. Barabasi and R. Albert. 1999. Emergence of Scaling in Random Networks. Science 286, 5439 (1999), 509–512.

[14] A. Broder, R. Kumar, F. Maghoul, P. Raghavan, S. Rajagopalan, R. Stata, A. Tomkins, and J. Wiener. 2000. Graph Structure in the Web. In WWW.

[15] M. Faloutsos, P. Faloutsos, and C. Faloutsos. 1999. On Power-law Relationships of the Internet Topology. SIGCOMM Comput. Commun. Rev. 29, 4 (1999), 251–262.

[16] M. S. Granovetter. 1973. The Strength of Weak Ties. The American Journal of Sociology 78, 6 (1973), 1360–1380.

[17] J. Kleinberg and S. Lawrence. 2001. The Structure of the Web. Science 294, 5548 (2001), 1849–1850.

[18] R. Kumar, J. Novak, and A. Tomkins. 2006. Structure and Evolution of Online Social Networks. In KDD.

[19] F. Liljeros, C. R. Edling, L. A. N. Amaral, H. E. Stanley, and Y. Aberg. 2001. The Web of Human Sexual Contacts. Nature 411 (2001), 907–908.

[20] A. Mislove, H. S. Koppula, K. P. Gummadi, P. Druschel, and B. Bhattacharjee. 2008. Growth of the Flickr Social Network. In WOSN.

[21] A. Mislove, M. Marcon, K. P. Gummadi, P. Druschel, and B. Bhattacharjee. 2007. Measurement and Analysis of Online Social Networks. In SIGCOMM IMC.

[22] M. E. J. Newman. 2001. The Structure of Scientific Collaboration Networks. Proceedings of the National Academy of Sciences 98, 2 (2001), 404–409.

[23] G. Siganos, S. L. Tauro, and M. Faloutsos. 2006. Jellyfish: A Conceptual Model for the AS Internet Topology. Journal of Communications and Networks 8, 3 (2006), 339–350.

[24] 2018. Ethereum in BigQuery: a Public Dataset for Smart Contract Analytics. https://cloud.google.com/blog/products/data-analytics/ethereum-bigquerypublic-dataset-smart-contract-analytics.  Accessed: 2010-10-12.

[25] https://github.com/sgsourav/blockchain-network-analysis