Consensus at the Threshold

Quick Look at Miners and Validators Crossing The Merge

Consensus is the backbone of crypto. And so, not surprisingly, it is a topic of much discussion and debate. Ethereum after The Merge is no exception: There is concern that proof-of-stake (PoS) validators have become too concentrated in the hands of a few, and that some are censoring transactions.

This is not a new debate. Before The Merge, there was already concern about Ethereum miner concentration and censorship. And the issue is not unique to Ethereum either. 60% of all Bitcoin hashpower was once concentrated in China. In a sign of how quickly things can change in crypto, this distribution of Bitcoin mining radically changed in less than 2 years, with the US now dominant (over 35% at the time of writing).

Approximate hashrate change in Bitcoin network 2020–2021; by Takens Theorem + tools

In all of these debates there is legitimate concern. The goal of crypto is to create a neutral, censorship-resistant and decentralized system robust to attacks from even well-resourced actors. Importantly, even under conditions that seem sanguine (like, say, our friends are the popular block builder), a network is still subject to theoretical risk, of attack vectors not yet known. Any concentration makes such attacks more likely.

Let’s take a simple quantitative look at this concentration of miners and validators before and after The Merge.

Crossing The Merge Threshold

Taking thousands of blocks before and after The Merge, we can look at the concentration of blocks for each associated block proposer’s address (miners and validators). Before The Merge, only a few dozen mining addresses were associated with rewards, which included Ethermine and others. After The Merge, thousands of addresses are associated with validation¹. The difference in the number of unique reward-recipient addresses is substantial, a sharp threshold crossing:

Data from Etherscan; plot by Takens Theorem

This is a promising feature of PoS. There is a 32 ETH staking requirement to participate as a validating node, yet it is low enough to have thousands of participating entities.

Still, the concentration of reward recipients remains quite high after The Merge. I can compute a windowed Gini analysis, a metric that reflects how concentrated each 1000-block window is. Gini varies from 0 (for perfectly shared blocks) to 1 (for a single validator taking them all). We can also calculate what percentage of blocks is taken by the top 3 associated with addresses. The two lines in the next plot show this analysis over time:

Data from Etherscan; plot by Takens Theorem

Gini is an imperfect measure, and there are alternatives and reasons for skepticism. But it is a canary of sorts². While there are more validators, there is also more concentration in (proportionally) fewer of them. Although the percentage control by the top 3 seems similar, there is a statistically significant difference between the left and right sides of this plot, of about 2% more concentration after The Merge (p < .005). Despite the potential for decentralization, there is considerable concentration under both metrics.

This concentration may be inevitable. The top 5 validating pools own almost 80% of staked Ethereum. And PoS is highly mobile. A validator can build their own blocks from anywhere, but also instantaneously switch to bids from other block-building services. It is a hallmark of the PoS shift: a further decoupling between block proposers and builders.

How can we ensure that centralization of block-building services does not unduly affect operation of the network? Jon Charbonneau at Delphi has a great, extensive analysis of this decoupling and how it could be optimized in the protocol. One way to do so, as Vitalik Buterin summarized in a recent post, is to optimize a balance between constraint on block builders and lower burden for the block proposer.

Vitalik considers several designs for striking this balance, such as “Inclusion Lists”

Censorship Issues

This concern about concentration is made worse by a trend observed by many after The Merge: Some block builders are still avoiding Tornado transactions. Indeed, Flashbots, the largest and most successful block builder, avoids Tornado transactions completely at the time of this writing. This represents a drawdown for these privacy-preserving transactions of somewhere between 30% to 40% of blocks. A key dispute in this issue is whether avoiding blocks with OFAC-sanctioned addresses is required. There has been no such specific advisement from the US government. Though it is quite rational to abide by local jurisdictions, the fuzziness of the situation causes some debate.

Using data from Etherscan’s dashboard tools, here is a recent breakdown of validators by percentage of blocks produced and percentage of Tornado Router (0xd90e2f9) transactions proposed. You can see Flashbots on the far left, the largest block builder, validating 0% Tornado transactions. Others may also be avoiding them (or using Flashbot’s relayer).

Data and plot with Etherscan tools; query by Takens Theorem

Charbonneau highlights the importance of this issue in his Delphi analysis, and doesn’t mince words: “Scaling and the like can honestly take a back seat though. Ethereum can survive with a bit higher fees, but it cannot survive with censorship.” One existing tool the network has for this issue is slashing. Validators could punish others for these kinds of practices. Slashing now happens occasionally on the network. It has been offered by Eric Wall and others as part of a solution.

Conclusion: Decentralization

As noted above, PoS introduces considerable mobility in the consensus process that GPUs and proof of work did not before The Merge. But mobility is a double-edged sword too: If you can move to decentralization quickly, you can also move quickly in the reverse direction.

One way to think about these highly “mobile” validating networks is that they have two types of decentralization: potential vs. realized. All networks have this property, but PoS may have a higher rate of change on this gradient—a potential to move more quickly between extremes. In other words, validators can easily move their operation both physically and across block-building services. While the network of physical validators and corresponding nodes could remain the same, the network’s wiring could in theory centralize very quickly: from independent block building to relying on a centralized builder.

Ethereum now has hundreds of thousands of unique validator addresses. This represents significant overlapping entities (“Sybil” issues; see Etherscan’s dashboard for some statistics on depositors), but it still means that Ethereum’s PoS has extremely high potential decentralization. If all depositors ran their own nodes, there might be many thousands of independent, unique entities proposing blocks.

But its decoupling of proposers and builders brings a second source of concentration: the block builders. A concern here is that there are incentives that hold between builders and proposers, because builders affect the profit margins of node operators³. That means that a highly successful builder can lower the decentralization of the network. Validators can flock to beautifully engineered block-building tools. As seen here in a plot of the past 2 weeks, Flashbots (red band) has come to dominate block building, representing over 30% just in its main fee-receiving address. This lowers Ethereum’s realized decentralization.

Data and plot with Etherscan tools; query by Takens Theorem

Kudos to Flashbots for sharing a transparency dashboard, where they report all of their relay and builder activity. (They’ve also just recently announced goals to share their infrastructure, create grants around transparency, and ultimately contribute to more decentralized block building.) At the time of this writing the total dominance of their service is at about 40%, and one can see why they are an attractive resource for validators:

From Flashbots transparency dashboard

I hope these ideas shared by Charbonneau, Buterin and from many other devs can further realize this great promise not just of energy efficiency and flexibility in protocol tuning — but also of massive potential decentralization of Ethereum after The Merge.

Follow me on Twitter. I was not paid for this blog post nor am I officially affiliated with any project mentioned here. This post is informational and not meant to be advice of any kind.

Endnotes

1. A very important caveat is that in both PoW and PoS Ethereum fee recipients do not always reflect total participants (such as in pools); I simplify for this analysis, but one response is to say that any such concentration is concerning.
2. Grim mining pun (for birds at least).
3. Charbonneau has a nice description of such a feedback loop.

Further reading

• Censorship… wat do? (Jon Charbonneau)
• Recent Ethresearch post by Vitalik Buterin
• Useful PoS notes from Kyle Waters of Coin Metrics
• Proof of Stake Is Less Wasteful (Eric Wall)
• The Case for Social Slashing (Eric Wall)
• If Ethereum Starts Slashing, It Burns (Nic Carter)