NNS and Decentralization
A review and analysis of the decentralization of one of the world’s largest protocol governance DAOs.
tl/dr
The NNS has evolved and continues to evolve in ways that increase the decentralization of NNS participants and, as such, reduces risks to the Internet Computer Protocol and its stakeholders. Decentralization of the NNS can be measured in multiple ways, including decentralization away from large Genesis token holders, voting power distribution and the ability for a future participant to capture significant voting share. In the first 21 months since the Internet Computer’s mainnet launch (Genesis) in May 2021 the majority of voting power has already shifted towards post-Genesis participants and the majority of ICP has already been staked on the NNS, practically eliminating the chance for a 51% attack.
1. Background: What is the NNS?
The Network Nervous System (NNS) is the permissionless on-chain DAO that allows the Internet Computer Protocol to be governed in an open, decentralized, and secure manner. The NNS has complete control over all aspects of the network, allowing it to be efficiently upgraded and expanded.
ICP is the utility token of the Internet Computer and token holders from around the world can stake their tokens within neurons in the NNS to engage in network governance and vote on whether to adopt or reject network proposals, while earning voting rewards for their participation. Anyone who stakes tokens in a neuron within the NNS can also submit a proposal to suggest changes, giving the ICP community the power to guide the blockchain’s evolution and optimize it for developers, entrepreneurs, and users alike.
Decentralization of the NNS is important to ensure that one entity, or multiple entities, can not collude to direct the Internet Computer in a direction that meets their self-interests at the expense of other stakeholders in the Internet Computer. This blog post is an analysis of the current status and trend of decentralization for the NNS.
2. Measures of Decentralization
This analysis will explore the progress and direction of the NNS towards decentralization across three measures:
- Genesis Entities — How has voting power shifted since Genesis?
- Nakamoto Coefficient — How much voting power is concentrated today?
- Future Participants and 51% Attack — How likely is voting power concentration to occur in the future?
- Liquid Democracy — How decentralized is the NNS liquid democracy system?
Note about analysis
Similar to Ethereum node operators and wallet addresses, it’s impossible to know the entities owning a neuron and whether an entity owns multiple neurons. In addition, public information on a neuron’s existence is only available if the neuron identification number is known. Identification numbers are typically made public (called “found neurons”) through the neuron owner searching for their neuron ID in the public dashboard. The exception to this is neurons created at Genesis, which are all found neurons.
Please note that “found neurons” is different from “known neurons”, which are neurons that show up in the NNS Frontend Dapp as options for users in the liquid democracy system (and were set by NNS proposals to self-identify and name the neuron).
Genesis Entities — How has voting power shifted since Genesis?
The Internet Computer publicly launched on May 10, 2021. During the Genesis launch, stakeholders in the Internet Computer project were granted ICP locked in neurons at different time frames. The owners of these neurons included the DFINITY Foundation, the ICA Foundation, seed-round and early contributors, strategic partners and team members. A full breakdown of the token distribution at Genesis can be found in a report by Messari. A key measure of decentralization is the movement of voting power away from these early stakeholders who tend to hold significant ICP due to their early contributions.
A combination of some Genesis neuron holders dissolving their neurons (and thereby being able to “un-stake” the neuron ICP) and new participants staking on the NNS has resulted in a majority of voting power being held by post-Genesis neurons (those created after Genesis). Post-Genesis neurons currently control 60.1% of the voting power on the NNS. DFINITY and the ICA foundations currently control 22.7% of voting power, a significant decrease compared to the 39.5% voting power they controlled just after Genesis. You can track DFINITY and ICA’s voting power on the public dashboard.
There are 4,592 seed and early contributor neurons that have voting power at this time. The largest of these neurons has 1.28M in voting power (0.29% of total voting power). Seed round contributors were given their initial ICP in 49 different neurons and the top 49 seed round neurons by voting power control 11.1M voting power (2.5% of total voting power). This means the top seed contributor could control, at most, 2.5% of total voting power, however it is likely that the top seed contributor actually controls far less.
There is currently 40.1M voting power (9.1% of total voting power) held by seed and early round stakeholders in neurons that are dissolving (the process of decreasing the amount of time for which the neuron must be staked and, at the same time, decreasing the voting power). As that 40.1M voting power decreases due to the neurons dissolving, the post-Genesis neurons are gaining relative voting power versus the seed and early round stakeholders.
Genesis Entities — Conclusion
In the first year and a half of the Internet Computer mainnet, the voting power held by Genesis neurons has decreased to 39.9% and will continue to decrease as some of their neurons dissolve and new participants stake on the NNS. This is a clear sign of decentralization away from the early stakeholders who tended to have large stakes and towards new participants in the Internet Computer governance, who tend to have smaller stakes.
Nakamoto Coefficient — How much voting power is concentrated today?
The Nakamoto Coefficient is the minimum number of entities needed to achieve 51% of voting power within a subsystem (a self-contained system of control) of a blockchain. This is calculated by summing up the largest entities of voting power until 51% of total voting power is reached. In the case of the NNS, 225M voting power is currently needed in order to have 51% of the total voting power. The below analysis focuses specifically on the NNS voting power subsystem and later articles will address the Nakamoto Coefficient for other Internet Computer subsystems like nodes and node providers.
It is only possible to perform this analysis on found neurons, which make up 64.6% of the voting power on the NNS. DFINITY and the ICA Foundations control a total of 22.7% of the found voting power. The remaining 41.9% of found neuron voting power is composed of 11,264 neurons. The voting power of those 11,264 neurons are broken down to 56.5% for seed round stakeholders, 23.2% for early contributor stakeholders, 19.4% for post-Genesis participants and 0.9% for node providers.
In terms of the Nakamoto Coefficient, DFINITY/ICA neurons plus the top 1,283 found neurons in voting power achieves a voting power of 51%. A worst-case scenario for the Nakamoto Coefficient for found neurons would be DFINITY/ICA plus 12 seed stakeholder entities (a single person or organization that owns or controls one or more neurons), plus 17 early contributor entities plus 17 node provider entities plus 87 post-Genesis participants, resulting in a Nakamoto Coefficient of 134. This scenario assumes that the largest neurons in the post-Genesis participants group are owned by different entities. For reference, Balaji Srinivasan, creator of the Nakamoto Coefficient, calculated coefficient values of 5 and 2 for main client code updates for Bitcoin and Ethereum in 2017. A more recent benchmarking can be found in a mid-2022 report from the Solana Foundation that calculated Solana’s Nakamoto Coefficient at 31, the highest of their Proof-of-Stake companions, however also consider that the methodology for updating mainnet code varies between PoS blockchains so this benchmark is not necessarily apples-to-apples. As mentioned above, a Nakamoto Coefficient is calculated for a subsystem. Additional articles will follow that calculate the Nakamoto Coefficient for additional subsystems, including nodes and node providers.
At the time of writing, the ICA dashboard reports 153,215 total neurons, 18,279 of which are found. The remaining 134,936 unfound neurons total 156M voting power, an average of 1,157 VP per neuron. One counter-argument to the above analysis is that if all of the 134,936 unfound neurons were owned by the same entity, then those neurons’ voting power combined with DFINITY/ICA’s voting power would be 58%, which would make the NNS’s Nakamoto Coefficient equal to 2.
Nakamoto Coefficient — Conclusion
Using data on found neurons, a worst-case Nakamoto Coefficient for the NNS is 134, which means it would take a minimum of 134 different people and organizations to collude in order to take actions against the best interests of the Internet Computer Protocol. This Nakamoto Coefficient for the NNS subsystem is relatively strong compared to other blockchain coefficients, particularly in subsystems that update main client code.
Future Participants and Sybil Attacks — How likely is voting power concentration to occur in the future?
At the time of this writing, it is impossible for a new participant in the NNS to acquire 50% of the voting power, regardless of their financial resources. This is because a new participant would need 220.75M ICP staked in an 8-year neuron to reach 50% of the voting power and there currently exists only 178.7M ICP not already staked on the NNS. In addition, with the exception of DFINITY/ICA, no current NNS participant could acquire enough ICP to achieve 50% voting power given that the maximum possible voting power for a Genesis entity is 11.1M (from the Genesis Entities analysis).
DFINITY/ICA could acquire 50% of the voting power, however they would need to stake an additional 120M ICP in an 8-year neuron in order to achieve this. This is theoretically feasible given that 178.7M ICP is currently unstaked, however it is unlikely. One reference point is there exists only 52.4M ICP in major exchanges.
Voting power concentration is extremely unlikely within a single entity due to the fact that the majority of ICP is already staked on the NNS.
Liquid Democracy — How decentralized is the NNS following system?
The NNS utilizes a liquid democracy system in which neurons have the option, but not the obligation, to delegate their voting power to one or more other neurons, through “following”. This is an extremely fluid system, as neuron holders can remove or change their following status at any moment. The NNS also has many categories of proposals (for example, ‘governance’ or ‘system canister maintenance’) and neurons can configure their following by proposal topic.
Recent NNS proposals show that over 99% of voting power is following the DFINITY or ICA neurons for non-governance topics. This has remained pretty consistent since Genesis.
For the governance topic (and for a new topic called SNS & Community Fund), the following structure is much more decentralized and distributed. It’s challenging to put exact numbers on follower counts since neurons can follow multiple neurons and can change their following election at any time, but an analysis of a recent governance proposal can provide insight to the voting power consolidation of the liquid democracy mechanism. This analysis used proposal 96475, for which 221M votes were cast. The top vote changes for that proposal by vote share of total NNS voting power were:
- 23% (DFINITY/ICA’s neurons, suggesting few neurons are following DFINITY)
- 5.6% (most likely the Synapse known neuron)
- 5.1% (most likely the ICDevs known neuron)
- 3.5%
- 1.3%
- 1.3%
The rest of the vote changes were less than 1% of total voting power. The vote change can be viewed on the proposal 96745 dashboard page. The proposal garnered 51% of total voting power.
One common viewpoint of liquid democracy is that the act of one neuron following another is analogous to a neuron casting its vote based on the reputation and recommendation of another neuron. In this viewpoint, the accumulation of voting power due to the liquid democracy system on the NNS does not cause centralization any more than a neuron participant seeking expert opinion on a proposal and voting based on that opinion. In addition, because a neuron owner has the ability to modify or remove their following status at any moment, risks of centralization from the liquid democracy system are reduced. Under this viewpoint the accumulation of voting power through liquid democracy is not contradictory to the goals of voting power decentralization.
Liquid Democracy — Conclusion
DFINITY’s 23% of voting power through following is greater than the voting power through following of the next five known neurons. This can change over time and the trend since Genesis has been towards voting power distributions across greater numbers of neurons and known neurons. One key area for improvement for Internet Computer governance decentralization is more participation in governance topics as recent proposals have had ~51% participation rates.
3. Overall Conclusions
The voting power on the NNS currently has a Nakamoto Coefficient of 134 for found neurons (64.6% of voting power), meaning that a minimum of 134 entities would need to collude in order to take actions that benefit themselves at the expense of other Internet Computer stakeholders. These 134 entities include the DFINITY and ICA foundations and a few seed round and early contributor stakeholders that are found and publicly known, but the majority of the 134 entities are not known. The NNS’s Nakamoto Coefficient has trended towards increasing over time as larger stakeholders dissolve their neurons and new, relatively-smaller participants stake on the NNS. Given that the majority of ICP is already staked on the NNS, it’s difficult to see how an entity could achieve a 51% attack without significant cooperation from many other entities.
The natural conclusion is that the risks associated with centralization of the NNS have significantly declined since Genesis and it appears it will continue to decline moving forward.
