Why the Blockchain Might Not Be As Decentralized as You Think

Table of Links

Abstract and 1. Introduction

2 Methodology

3 Hardware

4 Software

5 Network

6 Consensus

7 Cryptocurrency Economics

8 Client API

9 Governance

10 Geography

11 Case Studies

12 Discussion and References

A. Decentralization and Policymaking

B. Software Testing

C. Brief Evaluations per Layer

D. Measuring decentralization

E. Fault Tolerance and Decentralization

Abstract. Decentralization has been touted as the principal security advantage which propelled blockchain systems at the forefront of developments in the financial technology space. Its exact semantics nevertheless remain highly contested and ambiguous, with proponents and critics disagreeing widely on the level of decentralization offered by existing systems. To address this, we put forth a systematization of the current landscape with respect to decentralization and we derive a methodology that can help direct future research towards defining and measuring decentralization. Our approach dissects blockchain systems into multiple layers, or strata, each possibly encapsulating multiple categories, and it enables a unified method for measuring decentralization in each one. Our layers are (1) hardware, (2) software, (3) network, (4) consensus, (5) economics (“tokenomics”), (6) client API, (7) governance, and (8) geography. Armed with this stratification, we examine for each layer which pertinent properties of distributed ledgers (safety, liveness, privacy, stability) can be at risk due to centralization and in what way. We also introduce a practical test, the “Minimum Decentralization Test” which can provide quick insights about the decentralization state of a blockchain system. To demonstrate how our stratified methodology can be used in practice, we apply it fully (layer by layer) to Bitcoin, and we provide examples of systems which comprise one or more “problematic” layers that cause them to fail the MDT. Our work highlights the challenges in measuring and achieving decentralization, and suggests various potential directions where future research is needed.

1 Introduction

Bitcoin [131], the first blockchain-based distributed ledger, [3] put forth a new paradigm, that inspired numerous systems to enhance and expand its model and thousands of applications to be built on them. Alongside, a research discipline emerged across cryptography, distributed systems, game theory and economics, to analyze the properties and capabilities of this paradigm-shifting protocol.

Bitcoin’s arguably most important contribution was offering a solution to the consensus problem [115,140] in an open setting. Contrary to classic protocols, cf. [77], Bitcoin participants are not known a priori; instead, the system only assumes a peer-to-peer (P2P) synchronous network and a public setup.[4] Bitcoin’s core security argument is that, if a majority of computational power acts honestly, the protocol solves the consensus problem and implements a distributed ledger, as shown formally in [78,139,79]. This, in conjunction with the premise that computational power is widely distributed over the network participants, gives rise to the “security via decentralization” proposition: the system has no single point of failure, as any network participant is individually too weak to influence the properties of the protocol, no matter how they behave. Intuitively, a high degree of decentralization suggests that the trust for safe system operation is spread across the largest possible set of parties.

The appeal of this narrative, and the emergence of ledgers like Ethereum with APIs of higher functionality, gave rise to various “Decentralized Finance” (DeFi) [177] applications. Such systems have drawn the attention of industry, governments, regulators, and banks worldwide. Nonetheless, there is no agreement as to whether blockchain systems are decentralized, or even what “decentralization” entails, despite it being a topic of interest for centuries and across different disciplines [16,168,94]. Proponents often tout the existence of diverse communities, wide geographical distribution, or a theoretical ability of open participation as evidence of decentralization [5]. Antagonists point to power concentration around a few entities when it comes to system maintenance, protocol upgrades, or wealth ownership [155]. Interestingly, both sides might be correct at the same time — to some extent. Blockchains may exhibit high levels of decentralization w.r.t. some aspects, but not others. Thus, the pertinent question is more nuanced than the simple binary one “is the system decentralized or not?” — we are interested to know to what degree and in which aspects the system is (de)centralized.

Another common fallacy is perceiving decentralization as a goal, instead of a means to an end, and equating it with security, stability, or even efficiency. In reality, decentralization guarantees none of these properties. It can be synergistic to them, but in practice centralized systems can be more secure and fail-safe than decentralized ones and vice versa, depending on the relevant threat model. Still, it can be argued that decentralization’s major advantage from a security perspective is related to the system’s resilience to single points of failure.

With this as a starting point, our work sets on exploring decentralization across different layers, or strata, of blockchain systems. In particular, we select layers that influence a distributed system’s security properties, e.g., privacy or fault tolerance. Thus, centralization in one of our layers points to the existence of a single point of failure for the system as a whole w.r.t. one of those properties.

Our systematization effort aims to inform users, practitioners, and researchers, and to support policymaking and law enforcement processes. Decentralization — or the lack of it — plays a major role in policy discussions and the debate over the regulation of blockchain systems. For example, to determine if a digital asset constitutes a security, and particularly an investment contract, the US Securities and Exchange Commission (SEC) focuses on whether asset owners expect to profit via the efforts of “active participants” (APs), e.g., promoters or sponsors [171] (see also Appendix A). If a system is deemed decentralized across all layers, in effect there is no AP that the system’s stakeholders rely on for profiting, so the underlying token would not be classified as a security under this criterion.

We note that many blockchain systems can be argued to have a potential for decentralization, due to their permissionless nature. Specifically, by allowing any party to join, they may find themselves in a decentralized state. Nonetheless, our work focuses on characterizing the decentralization of systems as manifested in specific points in time based on the engagement they attract, thus exploring to what degree these systems realize their decentralization potential in the real world, irrespective of whether they can be decentralized in theory.

Related Work. Various research works have addressed the decentralization — or lack thereof — of blockchain systems, from some particular perspective. The research of Zhou [187] and Cho [43] highlights the risk of centralization that arises in the context of hardware, when specialized equipment is used by system maintainers to create blocks. This tendency is also acknowledged in the work of Ekblaw et al. [62]. Choi et al. [44] and Reibel et al. [147] reveal high levels of similarity in the codebases of different blockchain projects, alluding to centralization around the software used in distributed ledgers. An empirical study by Azouvi et al. [11] also looks at software centralization within a single project, i.e., when few individuals undertake the majority of the development process. Neudecker et al. [134] identify several ways in which the underlying network of a distributed ledger can impact its overall degree of decentralization, while Apostolaki et al. [4] examine centralization on the level of Autonomous Systems (ASes) as an enabler of routing attacks on blockchains. A plethora of studies, such as those by Gencer et al. [82], Gervais et al. [83], Valdivia et al. [172] or Lin et al. [119], have focused on the decentralization of the consensus layer, by measuring the “mining power” ratio of a system’s block producers. Another blockchain dimension whose decentralization has been thoroughly studied is the one pertaining to the economics of cryptocurrencies — often termed tokenomics. Sai et al. [157], Cheng et al. [42] and Ron and Shamir [153] analyze the distribution of transactions and tokens across parties, while Moore and Christin [129] touch on the subject of secondary markets and the risk carried by their potential centralization. Chatzigiannis et al. [39] point out that most blockchain light client schemes are vulnerable to centralization because of their reliance on centralized servers or full nodes, a concern also shared by Moxie Marlinspike [123]. Gervais et al. [83] present examples of centralization from the space of blockchain governance, and particularly conflict resolution, while Azouvi et al. [11] complement this work with a more systematic exploration of the contributors behind improvement proposals and discussions. Various works, such as those of Mariem et al. [122] and Sun et al. [161], turn their attention to the geographic dispersion of participants and infrastructure within a blockchain ecosystem.

Despite the breadth and depth of the research around blockchain (de)centralization and its manifestations, there have been few efforts so far to generalize or systematize this knowledge. Sai et al. [156] offer a blockchain centralization taxonomy, based on an algorithmic literature review and expert interviews. They treat ledgers as multi-layer systems, capturing 13 aspects of centralization over 6 architectural layers: Application, Operational, Incentive, Consensus, Network, and Governance. However, their work neglects some components, such as software centralization (as identified in [44,147,11]), or geographic decentralization pertaining to layers other than the network (for example, the decentralization of consensus participants, as studied by Sun et al. [161]). More recently, Zhang et al. [184] propose a taxonomy around five facets of decentralization: Consensus, Network, Wealth, Governance, and Transactions. They focus primarily on transaction centralization (w.r.t. the distribution of transactions to users), which is mainly a measure of adoption and usage, rather than a dimension with security implications. Their systematization also does not account for several factors identified in previous research, including hardware [187,43], software [44,147,11], or geographic [122,161] decentralization. Last, there exist some studies that approach the topic of blockchain decentralization from different perspectives, e.g., economic or social [23,22]. Notably, while all these works offer ample information on blockchain decentralization, none of them propose a consistent methodology for determining the decentralization level across all relevant layers.

[3] for the rest of this work we use the terms “blockchain" and “distributed ledger" interchangeably, even though strictly speaking, the latter describes an objective while the former is a means to it.

[4] Bitcoin uses the following newspaper headline as the common setup string: “The Times 03/Jan/2009 Chancellor on brink of second bailout for banks.”