Source-linked AI summary
Decentralization in Bitcoin and Ethereum Networks
Adem Efe Gencer, Soumya Basu, Ittay Eyal, Robbert van Renesse, Emin Gün Sirer
TL;DR
Decentralization in cryptocurrencies has been studied less empirically than its importance to these systems suggests. This paper measures Bitcoin and Ethereum using adapted Internet techniques, direct observations, blockchain histories, and the Falcon relay network, finding that neither system is strictly better overall and identifying improvement directions.
Problem
The paper addresses limited measurement evidence about how decentralized Bitcoin and Ethereum are in practice, despite decentralization being part of their value proposition.
Method
The study combines adapted Internet measurement techniques, direct network measurements, blockchain histories, and the Falcon relay network to assess Bitcoin and Ethereum decentralization.
Results
Neither Bitcoin nor Ethereum has strictly better properties overall: Bitcoin has greater network capacity but more clustering, while Ethereum has lower mining-power utilization and less variable block rewards.
Takeaways & Limitations
The findings support concrete improvement directions, including increasing Bitcoin’s block size by 1.7x without reducing decentralization beyond its 2016 level and using a relay network to improve Ethereum’s mining-power utilization.
Takeaways & Limitations
Bandwidth results can lose accuracy when measurement-side bottlenecks, BMS traffic, or intentional remote-peer traffic shaping affect observations.
Abstract
from arXiv · showhide
Blockchain-based cryptocurrencies have demonstrated how to securely implement traditionally centralized systems, such as currencies, in a decentralized fashion. However, there have been few measurement studies on the level of decentralization they achieve in practice. We present a measurement study on various decentralization metrics of two of the leading cryptocurrencies with the largest market capitalization and user base, Bitcoin and Ethereum. We investigate the extent of decentralization by measuring the network resources of nodes and the interconnection among them, the protocol requirements affecting the operation of nodes, and the robustness of the two systems against attacks. In particular, we adapted existing internet measurement techniques and used the Falcon Relay Network as a novel measurement tool to obtain our data. We discovered that neither Bitcoin nor Ethereum has strictly better properties than the other. We also provide concrete suggestions for improving both systems.
1 Introduction
The paper measures decentralization in Bitcoin and Ethereum because important network-level assumptions have largely rested on limited empirical evidence. It combines adapted Internet measurement techniques, application-layer approaches, direct network measurements, the Falcon relay network, and blockchain histories to compare the systems.
- Motivation: Decentralization fragments control over the protocol, while miner and full-node decentralization affect censorship resistance, protocol analysis, and resilience to attacks.Full nodes relay blocks and transactions to miners and answer end-user queries about blockchain state.
- Motivation: The study addresses debates and protocol decisions that have often been based on assumptions rather than measurements of the underlying networks.
- Approach: The authors present a comprehensive measurement study of decentralization metrics in operational Bitcoin and Ethereum systems.The study examines network properties, protocol requirements, security, and client interactions.
- Approach: The methodology combines adapted Internet measurement techniques, direct measurements from multiple vantage points, a year-long Falcon deployment, and Bitcoin and Ethereum blockchain histories.Falcon was deployed across five continents to ferry Bitcoin blocks and observe pruned blocks.
- Scope: The paper comparatively evaluates Bitcoin and Ethereum across decentralization metrics and reports distinct strengths and weaknesses rather than a universally superior system.Reported findings include Bitcoin’s greater bandwidth capacity and clustering, Ethereum’s lower mining-power utilization, and different mining-reward volatility.
2 Bitcoin and Ethereum
Bitcoin and Ethereum both use Nakamoto-style consensus and peer-to-peer communication, but differ in protocol design, block production, networking, and block-selection mechanisms. Ethereum incorporates pruned blocks as rewarded uncles, whereas Bitcoin’s non-main-chain blocks are pruned.
- Shared foundations: Bitcoin and Ethereum use Nakamoto consensus to regulate transaction serialization, despite significant differences in APIs, abstractions, and wire protocols.
- Bitcoin: Bitcoin targets 10-minute block production with a 1 MB maximum block size and disseminates block and transaction announcements through peer-to-peer flooding.At measurement time, the last 100 blocks had a 0.99 MB median size and a 9.8-minute mean interval.
- Bitcoin: Bitcoin peers form point-to-point TCP links through a protocol-level three-way handshake that exchanges client state such as blockchain height and software version.
- Ethereum: Ethereum combines Nakamoto-inspired design with GHOST and rewards pruned blocks, called uncles, while targeting a 10-to-20-second block interval.Its block size is indirectly determined by a fluctuating execution fee called gas.
- Ethereum: Ethereum uses UDP-based Kademlia-inspired node discovery, while other peer-to-peer communication uses TCP with encrypted and authenticated messages.The wire protocol is poorly documented, so the study relies on client implementations and Ethereum wiki pages.
- Measurement infrastructure: The measurement infrastructure contains 18 globally distributed nodes.
3 Measurement Infrastructure
The study uses Blockchain Measurement System to measure public Bitcoin and Ethereum nodes from multiple vantage points, supplemented by a Falcon relay network for Bitcoin blocks. Its interpretations assume reachable public nodes represent the wider networks, although hidden nodes remain outside direct observation.
- Measurement system: Blockchain Measurement System runs experiments lasting from days to 12 months to examine peer relationships, node capabilities, and system-level security and fairness.
- Measurement system: BMS connects to each peer, collects measurements, and disconnects before proceeding to the next peer.
- Measurement scope: The study measures Bitcoin nodes over IPv4, IPv6, and Tor, but Ethereum nodes only over IPv4 because Ethereum lacked Tor nodes and BMS could not discover enough IPv6 nodes.
- Measurement infrastructure: Multiple geographically distributed vantage points estimate peer-to-peer latency, with 15 of 18 infrastructure nodes on PlanetLab and three in Cornell’s academic network.
- Measurement infrastructure: Provisioned-bandwidth measurements use well-resourced machines with high download capacity and sufficient disk space so the apparatus does not become the bottleneck.
- Scope and assumptions: Interpretations treat reachable public nodes as representative, although NATs, firewalls, and mining infrastructure can hide nodes from public observation.
- Falcon relay network: Bitcoin’s pruned blocks require direct miner connections because clients delete them without affecting blockchain correctness.
- Falcon relay network: Falcon uses cut-through routing to relay Bitcoin blocks worldwide and is directly connected to at least 36.4% of Bitcoin’s hashpower.Its miner connectivity provides observations of blocks absent from other well-connected nodes.
4 Measurements
Measurements compare Bitcoin and Ethereum across node bandwidth, network structure, mining concentration, mining power utilization, and fairness. The results show different decentralization trade-offs rather than a single network dominating every metric.
- 4.1 Provisioned Bandwidth: 1.7×: Bitcoin’s typical full-node bandwidth increased from 33 Mbit/s in 2016 to 56 Mbit/s in February 2017.The increase suggests block size could rise by a factor of 1.7 without exceeding Bitcoin’s 2016 de facto centralization level.
- 4.2 Network Structure: Bitcoin nodes are geographically closer and more often hosted in dedicated datacenters than Ethereum nodes.Only 13% of Ethereum latencies are under 100ms versus 46% for Bitcoin; dedicated-hosting autonomous systems contain 56% of Bitcoin nodes and 28% of Ethereum nodes.
- 4.3 Distribution of Mining Power: 61% of Ethereum’s weekly mining power was shared by three miners on average, while no Bitcoin entity exceeded 21%.The top four Bitcoin miners held more than 53% of average mining power, and both systems relied heavily on few mining entities.
- 4.3 Distribution of Mining Power: A Byzantine quorum system of size 20 could achieve better decentralization than proof-of-work mining at much lower resource cost.The authors identify permissionless consensus without such a high degree of centralization as an area requiring further research.
- 4.4 Mining Power Utilization: Ethereum’s mining power utilization dropped from 88% to 74% around several real-world events, including the DAO exploitation and network attacks.The study analyzed weekly and daily pruned blocks in Bitcoin and uncle blocks in Ethereum against main-chain blocks.
- 4.5 Fairness: Bitcoin and Ethereum fairness distributions differed significantly, with standard deviations of 1.72 and 0.25, despite similar means of 1.22 and 1.08.In both networks, the top four miners were generally more successful at appending blocks to the main chain.
5 Related Work
Prior measurement work largely focused on Bitcoin, examining propagation, connectivity, mining-power concentration, and overlay behavior. This paper extends comparable decentralization measurement to Ethereum and evaluates proposed approaches for reducing participation costs.
- Earlier studies found block propagation latency dominates for blocks below 20 KB, with the threshold later increasing to 80 KB.
- Around 100 Bitcoin nodes represent more than 75% of mining power and were conjectured to connect well to major mining pools.
- Prior work also examined Bitcoin peer discovery and deanonymization of peers in the overlay network.
- Sharding and authenticated data structures have been proposed to reduce node resource requirements and broaden participation.
- This study scientifically tests whether intuitions from blockchain explorers hold and performs a comparable study of Ethereum.
6 Conclusion
This study compares decentralization in Bitcoin and Ethereum using novel and established measurement techniques. It finds trade-offs rather than a universally superior system, with distinct network, mining, and scaling characteristics.
- Bitcoin has a higher-capacity network than Ethereum, but its nodes are more clustered and likely concentrated in datacenters.
- Bitcoin and Ethereum have fairly centralized mining processes, indicating that further research is needed to decentralize permissionless consensus protocols.
- Ethereum’s block rewards have less variance than Bitcoin’s, while its mining power utilization is lower, likely because of its high block frequency.
- Bitcoin can increase its block size by 1.7x without decreasing decentralization compared with 2016, while Ethereum would likely benefit from a relay network.