Read on Twitter
Yesterday I talked about the benefits of total decentralization. Today I'll talk about the benefits of deliberately engineered *bounded* centralization.
Thread:
Thread:
The best argument in favor of any centralization is often about concentrated incentive: if a million people each have a tiny stake in an outcome and only tiny ability to influence it, they won't care (or worse, are easily led astray by side incentives)...
... but if a single person has large stake+influence, they'll take time to learn and optimize and use economies of scale to make+execute good decisions.
The best arguments against centralization are fragility, risk of abuse of power, low "bus factor" etc etc you know the drill.
The best arguments against centralization are fragility, risk of abuse of power, low "bus factor" etc etc you know the drill.
Can we combine the two? Well, blockchains combine the two already. Consider the block production mechanism: basically, every 15 seconds we randomly select a dictator, who has total control of what txs they include, and in econ speak is the "full marginal claimant" of the rewards
This architecture actually has a lot of benefits, and is by far the most resilient of all architectures we know against bribes / "side payments".
Miners and transaction senders do sometimes make private deals with each other, but the benefit is low so it happens quite little.
Miners and transaction senders do sometimes make private deals with each other, but the benefit is low so it happens quite little.
And it has bounded downside: you don't like one block producer, wait 15 seconds for the next one.
Once there are more complex apps on chain, we see a challenge. Optimal tx inclusion, including capturing "miner extractable value" via frontrunning etc https://arxiv.org/abs/1904.05234 becomes complicated.
This introduces an edge for sophisticated block proposers, risking centralization.
This introduces an edge for sophisticated block proposers, risking centralization.
Here deliberately engineered bounded centralization comes to the rescue.
What if we split up consensus into two tasks:
1. Choosing what txs go into blocks (complex hence vulnerable to centralization)
2. Verifying others' blocks with txs (routine task hence more decentralizable)
What if we split up consensus into two tasks:
1. Choosing what txs go into blocks (complex hence vulnerable to centralization)
2. Verifying others' blocks with txs (routine task hence more decentralizable)
The current eth1 -> eth2 roadmap *kinda* has this: eth2 validators have to explicitly sign up to be "eth1 friendly" block proposers.
https://ethresear.ch/t/alternative-proposal-for-early-eth1-eth2-merge/6666
Proposal is more resource-intensive, validation less resource intensive, validators don't have to be proposers.
https://ethresear.ch/t/alternative-proposal-for-early-eth1-eth2-merge/6666
Proposal is more resource-intensive, validation less resource intensive, validators don't have to be proposers.
If proposers centralize, this poses much less risk to users than validators centralizing: even if 90% of proposers are bad, the worst they can do is delay transactions an average of 10 slots (~2 min). If they propose bad blocks, validators will ignore them.
Another even more explicit example of this is miner extractable value (MEV) auctions.
https://ethresear.ch/t/mev-auction-auctioning-transaction-ordering-rights-as-a-solution-to-miner-extractable-value/6788
Basically, in a rollup, auction off the right to aggregate and choose ordering of transactions to the highest bidder, eg. once per day.
https://ethresear.ch/t/mev-auction-auctioning-transaction-ordering-rights-as-a-solution-to-miner-extractable-value/6788
Basically, in a rollup, auction off the right to aggregate and choose ordering of transactions to the highest bidder, eg. once per day.
This completely separates the economies-of-scale-heavy centralizing task (tx selection) from the decentralized task (validation), and furthermore captures revenue!
And it's bounded risk, because anyone can publish a tx directly to chain and force inclusion in the next batch.
And it's bounded risk, because anyone can publish a tx directly to chain and force inclusion in the next batch.
Thus, mechanisms that concentrate and embrace economies of scale (and hence centralization) in one area can drain away economies of scale (and hence lead to more decentralization) in surrounding areas.
