Failure modes for decentralized systems

I believe any decentralized system will eventually tend towards one or more of these failure modes in which control of the system becomes, effectively, centralized:

(unfold the summary of each failure mode for more information about it)

a system which is essentially decentralized, but some vital piece or vital pieces of its infrastructure are designed in a centralized way

The Internet and its de facto dependence on DNS is the best example of this. The domain name system is centralized in two ways: first through the control of the root domain . being vested in thirteen root DNS server operators (which doesn’t seem to have had any ill effects yet, but it’s worth noting that the vast majority of those are directly subject to the jurisdiction of the government of one country, the United States); secondly through the centralized control of the top-level domains.

The latter has led to sites hosting content which powerful entities wish to censor, such as Sci-Hub and Library Genesis, having to jump between country-code TLDs on a regular basis, continually trying to find countries whose governments and designated registry operators won’t be bullied into revoking their DNS entry.

The massive expansion of the gTLD space was actually potentially an improvement in this situation, but the high cost of a gTLD means that those most threatened by the centralization of TLD control are shut out from the benefits of their own TLD.

a system which is essentially decentralized, but is dominated by a small number of incredibly large operators who together hold one single effective monopoly

This maybe doesn’t sound so bad, because in theory smaller players in the system can still exist. However, in practice, the interests of the largest operators tend to align to maintain their oligopoly, make it impossible for new players to join, and slowly force the remaining smaller ones to use their services.

Email is the best current example of this: the largest providers, Google and Microsoft, have biased anti-spam algorithms which more or less guarantee that any very small email server no longer has any chance at all of getting mail delivered reliably to their users. You can’t reasonably consider running an email server for your personal private mail any more, or for companies in the small business or Mittelstand category. Even institutions with many thousands of users, such as universities, are finding it difficult to guarantee delivery of email from their own servers, increasing the pressure on them to switch to Gmail or Microsoft Exchange.

The largest blockchain-based networks also seem to generally fall into this category.

a system which is in principle decentralized, but is almost entirely dominated by a single operator

Current examples of this failure mode in decentralized networks are rare, probably because of effective regulatory pressure in the specific case of networks which prevents actual monopolization. Note, however, the previous failure mode, where a duopoly, triopoly or similar becomes just as insidious as an actual monopoly, and regulators worldwide seem mostly blind to this variety of control so far.

The best historic example is likely the Bell System’s control of the North American telephone system. Telephone numbers and the telephone network itself are essentially decentralized, and networks interoperated almost from the very start: even at the Bell System’s peak, very small pockets of the US had managed to resist its monopoly (and once international lines became available, other countries’ networks of course remained free of it, even though they worked together technically). But the Bell System’s control of the vast majority of telephone networks significantly restricted what even the parts of the system beyond its direct control could practically be used for.

This is to be distinguished from market control by a single vendor, which creates the next failure mode.

a system which is decentralized, but has enforceable (and enforced) regulation under the central control of a specific group

The World Wide Web is currently the best example of this. The number of browsers in widespread use is so small (and, excluding a very brief period prior to the release of Netscape in the 1990s, has always been so small) that the control of the Web platform in practice belongs to the small number of (now mostly very large and powerful) entities which make those browsers. Browser vendors worked together to add DRM controls to the Web platform against the clear wishes of users and developers. Since Chrome has now reached the status of being used for the majority of Web traffic worldwide, Google in particular has an outsized voice in the process of deciding what the rules are for how websites are made.

Blockchain-based networks are also intrinsically examples of this, since, although the network itself is decentralized, the software for each network is invariably a monoculture, with only one implementation under the control of a small group of developers.