(Beta)

Centralization vs Decentralization

What are the common tradeoffs between centralization and decentralization? This is a common tension in political systems, economic systems, software systems, societal infrastructure, and the organization of institutions. Real world systems are often a blend, where some parts are centralized, some parts are decentralized, or parts have elements of both.

Even for decentralized systems, it's common for the design and control of the communication and coordination schemes to be centralized.

Centralization is not binary, it's a spectrum of concentration of authority. Most aspects of systems exist as a point on a relative spectrum of centralization. For example, an industry with 4 major cellular service providers is more centralized than one with 10 providers and more decentralized than an industry with 1 provider.

This topic is organized as a list of dimensions commonly traded off between centralization and decentralization. For most dimensions, sometimes a centralized system is better, sometimes a decentralized system is better. It depends on the specific system.

The point of this topic is to assist those who architect systems, and those who evaluate them, by providing judgement criteria.

In this text, "C" refers to Centralized and "D" refers to Decentralized.

Control — C concentrates control, D distributes it. Sometimes central control has benefits (ex: quicker decision-making). Sometimes decentralized control has tradeoffs (ex: checks and balances can result in balanced decisions and protection against extremism, but also result in gridlock).

Efficiency — C often makes more efficient use of resources, such as ___________. Sometimes C is less efficient, such as when the USSR attempted to centrally plan their economy. D is more efficient in the case of ______.

Performance — Sometimes C is more performant. For example, the speed of Credit Card transactions is faster than Bitcoin. Sometimes D is more performant, such as distributing files at scale via Bittorrent often delivers faster download speeds than a single server. When the performance of C is worse, it's often because of scability or resource limitations. When the performance of D is worse, it's often because of coordination costs.

Resilience — Resilience is always with respect to the kind of potential shock, and resilience may be measured in multiple ways — the likelihood of shocks, the severity of shocks, how much of the system is affected, and how quickly the system recovers. D is more resilient when shocks affect a part where they would have otherwise affected the whole of a C system. D is more resilient when recovery speed benefits from distributed, self-organized coordination. C is more resilient when they can enforce better protections, and where recovery speed benefits from quick decision-making.

Complexity — C is typically less complex. D typically requires additional coordination and may have duplication of resources.

Evolvability — Both C and D can be resistent to change (email protocol). When C is resistent, it's often because of disincentives. When D is resistent, it's often because of large coordination cost.

Fairness

Distribution of Costs & Benefits —

Variety & Customizability — D typically enables more variety and customizability.

Censorability — D is typically harder to censor.

Scalability — D is typically easier to scale.

Consistency — C usually provides more consistency, usually a result of central control and efficiency resulting in more homogeneity.

Accountability — C is typically more accountable. It's more practical to assign credit and blame for desirable and undesirable outcomes.

Security — C systems might be more secure because they can more easily control their parts and stay up to date on security best practices. However, C systems are often homogenous which means if they're compromised, they're entirely comprised. D systems that are heterogeneous may only suffer a partial breach.

Monitorability — D services are typically harder to monitor, which is often helpful to mitigate abuse. Tor and Bitcoin, for example, enable illegal activities that's difficult for law enforcement to track.

{TODO: this topic needs work! There are probably more dimensions to list, and those here need better explanations and more examples.}