Network consensus, whether based on PoW or PoS, must prevent Sybil attacks. This ensures that a single entity cannot create multiple fake identities to gain control over the network. Operating a full node costs only a few dollars per month and thus cannot guarantee security or decentralization. The number of block producers is crucial, as they play a far more significant role than regular full nodes. Furthermore, it is a high distribution of decision-making power and incentive model.
Where does the confusion come from?
Many people still believe that the number of full nodes determines decentralization and security. This misconception likely stems from Bitcoin’s early days when every new full node connected to the network genuinely increased decentralization and security. Back then, most full nodes actively participated in BTC mining, producing new blocks and earning rewards.
Two major developments changed this dynamic: the emergence of mining pools in 2010 and the rise of ASIC miners in 2013. Desktop computers and laptops running full nodes could not compete with the vastly superior computing power of ASIC miners.
As a result, block production became concentrated in a few nodes called pools, to which miners contributed their hash rate. Regular user-operated full nodes ceased to play a significant role in decentralization and network security, using only a tiny fraction of their CPU power.
In the left part of the picture, you can see the early days of Bitcoin with 6 mining full nodes. At that time, each user running a full node had control over the network. The right part of the picture shows the current distribution of decision-making power in the Bitcoin network. Now, only 2 full nodes (pools) produce blocks, supported by miners who delegate their hash rate to them. Full nodes no longer directly participate in block production; they passively accept new valid blocks or discard invalid ones.
What determines control over the network is computing power (CPU or ASIC), not a full node whose CPU is not used for mining.
It is important to note that in Bitcoin’s early days, all users who operated full nodes also controlled the network because they were simultaneously miners. This is no longer the case today. Control over the network has shifted to miners (and partly also the operators of pools), who are more like mercenaries than actual network users. As a result, the interests of users and miners may not always align.
Many people have not yet recognized this shift and continue to overestimate the importance of full nodes for blockchain networks. Running a full node is relatively inexpensive compared to the costs associated with mining. An individual can easily allocate many IP addresses and run numerous full nodes. However, full nodes lack protection against Sybil attacks, meaning they cannot significantly contribute to network decentralization—perhaps only minimally in terms of governance.
What truly matters for the decentralization and security of blockchains is the distribution of expensive resources and the incentive model. In Proof-of-Work (PoW) networks like Bitcoin, this means the distribution of hash rate. In the case of Proof-of-Stake (PoS) networks like Cardano, it involves the distribution of ADA.
PoW miners and ADA stakers do not need to run a full node to wield decision-making power within the network.
How important are full nodes?
A full node has the capability to verify the entire blockchain history, allowing each user to independently ensure the network’s integrity. While this feature is crucial for blockchain security, most people rely on third parties.
Approximately 90% of users store their cryptocurrencies on centralized exchanges. Very few self-custody users operate a full node. I don’t know of any network where more than 0.001% of coin holders are running a full node at the same time. The full nodes run by block producers and exchanges are typically the most critical for the network, as their services are utilized by the vast majority of users.
If the team calls for a network upgrade that is not backward compatible, it’s crucial to wait until the majority of full-node operators have installed the new version of the client before initializing the new functionality. The priority is to ensure that block producers, centralized exchanges, and operators of key infrastructure nodes are upgraded first. Full nodes operated by regular users are less critical and can be upgraded at a later stage.
From the perspective of network consensus, regular full nodes are the least important, and their role is largely negligible. They passively consume blocks produced by block producers’ nodes and lack mechanisms to alert or correct the bad behavior of block producers.
Full nodes operated by block producers actively participate in producing new blocks and must have compatible client versions. It is crucial for them that new blocks are accepted by other block producers, as they rely on subsequent blocks being appended after theirs. Block producers and delegators can continuously add new blocks without needing to communicate with regular full nodes or care about their opinions. Hypothetically, even if all regular full nodes suddenly disappeared or stopped accepting blocks for any reason, block producers would still be able to continue building the blockchain.
Of course, it is necessary to consider the social dynamics between block producers, delegators, full node operators, and users. Conflicts between block producers and regular full node operators would attract community attention. Block producers cannot ignore the fact that some full nodes are discarding the blocks they consider valid. Such conflicts would need to be resolved because the network must ultimately serve its users. Block producers rely on user support and are economically incentivized by transaction fees. However, as mentioned, only a small portion of users run full nodes. This gives block producers a very strong position in the network, which they could potentially abuse.
For instance, if a pool censors transactions from a significant number of users or participates in MEV attacks, it can still produce valid blocks that regular full nodes will accept. While full node operators might learn about these issues from other sources, such as mem-pool analysis, their full nodes won’t assist in addressing the problem.
While block producers can profit from MEV attacks, regular full nodes cannot and are powerless to prevent this malicious behavior. They only accept valid blocks AFTER they have been produced.
If a full node operator refuses to accept valid blocks due to MEV attacks, their version of the blockchain would fall out of sync with that of the block producers. Full node operators must accept the execution of MEV attacks without the ability to protest, placing them in a weaker position within the network.
The network is controlled by those who hold expensive resources—let’s call them delegators. These delegators assign hash rate or PoS coins to pools, allowing them to enforce accountability by switching their delegation to another pool that behaves fairly. Notably, operating a full node is not necessary for this enforcement.
This is why decentralization should be measured by the number of block producers and delegators, along with the distribution of expensive resources among delegators. Delegators are not equal, as more money can buy more resources and thus more control. The number of regular full nodes is irrelevant; they serve only as passive monitors of block production.
Consider a more extreme scenario where a dominant portion of delegators and several pools start producing incompatible blocks. Suppose these pools install a modified version of the client that is not compatible with the version running on most full nodes in the network. This new client version could include a feature that some delegators and pools want to forcibly implement, or it might enable some form of fraud.
If a significant share of the network’s resources participates in such an attack, full node operators would be powerless to actively defend against it. This situation would lead to a blockchain fork and ensuing chaos. Under the Nakamoto consensus, used by both Cardano and Bitcoin, the longest chain rule applies. However, this rule would result in different chains prevailing on different nodes, leading to fragmentation and uncertainty.
Fraudulent pools, supported by a dominant share of the expensive resource, would produce a longer chain. Full nodes running the original (older) client version would reject the new blocks from these fraudulent pools and only accept blocks from pools with a minority share of the expensive resource. As a result, blocks on each competing chain would be created at a slower speed, but the fraudulent chain would ultimately be longer due to slightly greater support from the holders of the expensive resource.
How would full-node operators react? They would face a choice: remain on the old client version and accept a shorter chain with less resource participation, or adopt the new client version and follow the longer chain rule.
Predicting the outcome of such an attack is challenging, as it would depend on various factors, such as the actions of centralized exchanges, the stance of the core development team, and the influence of prominent figures in the community. In Bitcoin’s case, ETF issuers would also have significant influence.
The number of full nodes in the network is irrelevant to both the initiation and the outcome of such an attack. Each full node operator must make an individual decision, but this decision will be heavily influenced by the actions of other key players.
If other pools in the network installed the new version of the client and the delegators remained loyal to these pools, support for the old version of the client would gradually fade to the point of complete extinction. This example illustrates well how much more important nodes of block producers are compared to regular full nodes. The pressure of users and investors could have a stronger weight than the pressure of full-node operators.
The distribution of the expensive resource among delegators and the number of block producers are more critical for security and decentralization than the number of full nodes.
The greater the number of block producers, the harder it is for them to collude on an attack. This difficulty also extends to delegators. While an agreement among pools holding a significant share of the expensive resource may be sufficient to initiate an attack, it may not be enough to sustain it. Some delegators might only notice the attack after it has begun. Once they evaluate the situation, they are likely to redelegate their stakes or hash rate to other pools, thereby weakening the attack.
Conclusion
A large number of full nodes can indicate the health and robustness of the network’s infrastructure. More independent full nodes monitoring block production and network events contribute to this health. In Satoshi’s ideal world, every user would run a full node and have control over the network. However, today we know this is unrealistic. For most users, even self-custody is a significant hurdle, let alone running their full node.
From the perspective of decentralization and security, the number of full nodes is insignificant. The main reason is that running a full node has no protection against Sybil attacks, and full node operators have low skin in the game compared to mining or holding coins.
The comparison between hash rate and ADA coins is interesting. It is costly and impractical for most BTC holders to run an ASIC miner. They hold BTC as a valuable asset but still have no control over the Bitcoin network other than the ability to sell their coins if they don’t like something. ADA holders, on the other hand, have direct control over the network. If a pool censors transactions or conducts MEV attacks, ADA holders can delegate their stake elsewhere. This means that network users have control over the infrastructure they use, unlike BTC holders who must rely on pool operators and miners.