Centralised vs. Decentralised vs. Distributed: Clearing up a Messy Conversation

People keep using centralized, decentralized, and distributed as if they’re interchangeable. They aren’t. The confusion leads to bad expectations, sloppy architecture debates, and persistent myths about networks like Mastodon.

This post clears up the terminology with practical, real-world definitions and explains why federation is not the same thing as distribution.

Why This Matters

These words shape how systems behave, fail, and evolve. They determine:

• Who holds authority
• Where trust lives
• How failures propagate
• What resilience actually looks like
• How users should expect data to behave

And when we blur the concepts, we misjudge the systems in front of us. You can see this clearly in social platforms where many insist that something like Mastodon is “distributed,” when it is very much not.

Centralised (Hub)

A centralized system has a single authority. Everything—identity, policy, data—flows through that core.

Characteristics:

• One authoritative hub
• Clear governance
• Efficient to operate
• Fragile: one outage or one bad actor can take down (or distort) the entire system

Examples:

• Facebook / Instragram / Twitter / X
• Gmail / Yahoo Mail / Outlook / iCloud
• Traditional SaaS platforms

Centralization is simple and predictable. It’s also brittle and trust-heavy.

Decentralized (Federated)

Decentralization spreads control across many independent authorities. Each node—often a server or operator—governs its own domain.

This describes the ActivityPub ecosystem and services like Mastodon.

Characteristics:

• Multiple independent servers, each with local authority
• No global controlling entity
• Interoperability via standards
• Local failures are isolated
• Users still depend on the reliability and integrity of their chosen server

Examples:

• Email
• Mastodon
• Most ActivityPub-based systems
• DNS (operationally decentralized, despite its hierarchy)

This model offers stronger resilience than centralization but still creates anchor points. Pick a bad server, and you inherit its problems.

Distributed (Peer-to-Peer)

A distributed system spreads computation, state, and responsibility across a peer network. Nodes coordinate algorithmically—usually without privileged authorities.

This is where most misuse happens. People see “servers talking to each other” and assume distribution. True distributed systems behave very differently.

Characteristics:

• No single authoritative operator
• Peers replicate or share state
• Consensus or coordination protocols replace institutional trust
• High fault-tolerance by design
• Identity and data don’t hinge on any one node

Examples:

• BitTorrent
• Bitcoin, Ethereum
• IPFS
• CRDT-based peer networks
• DHTs

Distributed systems aim to resolve authority via algorithms rather than manual policy decisions. Administrators can still tweak those algorithms, but the system’s coordination is always governed by the protocol.

Where the Confusion Comes From

Three drivers cause people to conflate decentralization with distribution:

1. Marketing language: “Decentralized” became a buzzword long before people clarified what it meant.

2. The false binary: Users tend to think: “It’s not centralized, so it must be distributed.” They forget that federation—decentralization—is its own topology.

3. Peer illusion: Federation looks like peers exchanging data. In reality, servers are independent authorities, not algorithmic equals.

Mastodon Is Decentralized, Not Distributed

A Mastodon server controls:

• identity
• data
• user experience
• moderation policy
• availability

If that server fails, users tied to it suffer. That alone disqualifies the system from being distributed, because truly distributed systems replicate identity and state across many peers so that a single node going offline doesn’t wipe out anyone’s presence or data.

Mastodon’s architecture is federation, not peer-to-peer distribution. Servers don’t replicate or share global state, and no consensus protocol coordinates the network.

It’s decentralized, and that’s fine. It just isn’t distributed.

Practical Mental Models

Centralized

1           [ Hub ]
2          /   |   \
3        User  User  User

Decentralized

1    [Server A]   [Server B]   [Server C]
2       /  \         /  \         |
3    User  User   User  User    User

Distributed

1      Node — Node — Node
2        \     |     /
3         Node — Node

These diagrams eliminate most misunderstandings immediately.

Terminology Snapshot

A quick glossary keeps the rest of the post grounded in the terms we’re really trying to compare.

• Centralized networks: One operator owns the identity, policy, and data flows. The Hub diagram above is the clearest analogy—you trust the hub, and when it fails the whole experience collapses.
• Distributed networks: Nodes replicate or agree on shared state via algorithms or consensus; no human-run authority dictates behavior. The Peer-to-Peer diagram shows the mesh where peers coordinate directly.
• Decentralized networks (federation): Independent servers (like Mastodon instances) govern their own users but interoperate through standards. That matches the Server A / Server B diagram and is what most people mean when they say “federated.”
• Decentralized ecosystems (client-driven): There is no federation layer—users host their own feeds (Twtxt, Yarn.social, POSDE-style publishing) and clients pull via conventions. Authority is still distributed, but there are no server-to-server exchanges or replicated global state.

Each row in the glossary maps back to the illustrations above, so you can see how authority, communication, and resilience change with every topology.

Why the Distinction Matters in Real Systems

Trust Model

• Centralized: trust the operator
• Decentralized: trust your operator
• Distributed: trust the protocol

Resilience

• Centralized: vulnerable to collapse
• Decentralized: resilient to local outages
• Distributed: designed to survive massive node failures

Governance

• Centralized: organizational
• Decentralized: local and cultural
• Distributed: embedded in algorithms and consensus rules

If you mislabel the system, you misjudge the risks.

Closing Thoughts

“Decentralized” and “distributed” are not interchangeable. Federation is not peer-to-peer. And systems like Mastodon are not as resilient—or as trustless—as many assume.

Getting the language right helps us build better networks, set accurate expectations, and avoid overhyping architectures that don’t match their marketing.