These are some incomplete notes from the latest debate between Anatoly Yakovenko and Justin Drake. I covered key topics that I found interesting such as timing games, issuance as a cost, and economic security. The debate also triggered some discussions around CT so I also quoted some tweets that are worth reading.
Timing Games
Solana has 400ms block times. JD argues that MEV grows linearly with time, so shorter block times, as in Solana, could lead to more MEV opportunities. Even a few milliseconds advantage can give you an edge on MEV strategies. In a high throughput network, a small latency advantage could result in more txs landed on your blocks. This could potentially result in increased income and more delegates for your validator. Validators with an edge in terms of latency will make more money as they’ll produce blocks with more MEV.
If Solana block time is 400 ms, and you have a 200ms edge at best, then you get 50% more time to build your blocks, and you get 50% more txs landed, and therefore more ROI.
Timing games are not new, they have been known for a while. The strategy is to delay block transmission until the last possible moment of your turn as a block producer, so you can maximize your ROI by the number of txs that you listen on your turn, extracting more MEV.
The downside of it is that you’re playing at the limit, it can also happen that you fail to transmit your block on time, so you’ll be overrun by the next validator in line.
https://x.com/casparschwa/status/1660931488110583811?s=20
These timing games are not just a centralization vector, as they elevate the barrier to entry; In order to even be able to play the game, validators need to have outstanding infrastructure, even above the avg infra in Solana, but these strategies also degrade network consensus as delayed block proposals may lead to more missed slots, reorgs, and incorrect attestations.
Toly disagrees with this PoV. He argues that a lot of txs land before a validator’s slot begins, so the extra time during the slot accounts for a small % of total txs. He doesn’t see this as a realistic outcome, pointing out the difference in APY between MEV-boost client and vanilla validators in Ethereum as an example, which is not as large as JD suggested. He’s actually right, as evidence shows MEV-boost vs vanilla staker APY is not that different (~3.4% vs ~3.1%). Average Solana validators’ APY is ~7.4%, whereas Jito’s validators are earning ~8.14% (10% more than average).
https://twitter.com/uriklarman/status/1795562026703478789
He mentions that the median transaction fee in Solana is quite low (~0.2 cents), so the additional income from having an edge in block times would be insignificant, the extra txs that come in during the slot time are not necessarily higher in value.
Personally, I think if a validator realizes that it can gain an edge (even the smallest one), they’ll exploit it to the hilt. While it might not be profitable to play the timing games now, it doesn’t mean it won’t be profitable in the future. Every L1 will follow the natural path that Ethereum laid out: while ago, Solana didn’t have an MEV client and their spam problem was way worse than it is now. Then, Jito built it. And now they’re looking to implement PBS.
If Solana moves towards implementing PBS, I definitely think we’ll see builders playing timing games as the landscape would change by decoupling block proposers and builders by auctioning off the right to build a block. Let me explain:
Currently, I think incentives for delaying block submission by block producers is not enough as they risk their slot being missed, thus risking rewards. In PBS, builders are well incentivized and it’s more rational for them to try to maximize block profits for block producers to choose them over their other builders, and gaining ~10% more it could mean a huge advantage.
He further explains that Solana’s design provides guarantees that when a validator votes on a block, the vast majority of the network has also received that block, which negates the advantage of having a faster connection. Toly states that Solana’s goal is to solve these problems with hardware and create a competitive environment.
As Sreeram stated, I also think they both are right in a way: shorter block times in Solana means less MEV as validators don’t have much time to include txs on their turn, but this also translated it into more timing games as every validator seeks to maximize profits but only a few will have the expertise to do so.
https://x.com/sreeramkannan/status/1797834165926412444
Towards the end, both of them made two predictions:
- JD said that short block times on Solana could lead to hyper centralization within a given space, and that it places incentives on co-location to maximize validators’ ROI.
- Toly said that if the parameters are set correctly, a validator anywhere in the world could vote and achieve similar APY as anyone else, minimizing the impact of latency advantage.
Is Issuance a Cost?
The starting point of the debate was Toly claiming “issuance is not a cost to the network”.
Justin made the argument of issuance being a cost by explaining that for paying taxes, stakers would need to sell their earnings, so this selling pressure was presented as a cost to the network.
Toly compared selling Apple’ shares to the above example, arguing that selling stocks is not a cost to Apple. I’m not sure that I agree with his rationale, as Apple does not pay his employees by issuing Apple stocks (not counting bonus and stuff). ETH stakers’ earnings are paid entirely in ETH. Therefore, their profit depends on selling it to cover living expenses, not just taxes.
https://twitter.com/jon_charb/status/1798018320232489007
I like the point that Jon made here. In the end, it all comes down to a semantic argument, relative to who is the viewer: non-stakers are diluted at the expense of stakers; stakers’ income is a cost to non-stakers. Or saying it in another way:
- For non-stakers, issuance is a cost
- For stakers, issuance is their income.
Also a good reading – https://twitter.com/weboftrees/status/1797793469324476761
Economic Security
Toly believes economic security is a meme and that token issuance does not provide economic security. He claims that security of the chain is based on physical infrastructure and geographically node distribution, and that double spending can be prevented through a decentralized network and invalid state transitions can be prevented by filtering out invalid blocks with full nodes. By running a full node, you just filter invalid blocks out. Says that Solana testnet and even Chihuahua chain can provide an equivalent amount of security than Ethereum Mainnet.
To perform a 51%, the attacker needs to go buy a huge chunk of tokens, that will be detected. Once it’s detected, the attacker is slashed by off-chain coordination.
The fact that it’s detectable and easy to slash are the things that secure the Solana testnet and the tendermint Chihuahua chain. And this is why you don’t see these attacks anywhere
Toly argues that the social layer can intervene and restore the chain within 24 hours. However, it’s important to note that we’re potentially discussing an event that could take days to resolve. The time required to reach an agreement among all involved parties could significantly exceed 24 hours.
JD claims that we never saw this attack because two things need to happen:
- A sophisticated attacker who is aware of this attack and is willing to execute it, and
- The incentives need to be there. Right now, almost all of the value secured on Ethereum is ETH itself, and it’s staking. What will happen when there’s trillions in DeFi protocols and in stablecoins?
https://x.com/sreeramkannan/status/1797834165926412444
Sreeram made a good thread, I’ll stay with this phrase: “for these defi protocols to be economically safe, the total stake should exceed the total amount on these protocols”.
Once the incentives to, for example, censor oracles or frontrun liquidations are properly aligned, then you’re in a position where you can harvest toxic MEV. And if the amount of toxic MEV that you can harvest is greater than the cost of attack, then it’s going to happen.
https://twitter.com/aeyakovenko/status/1797757712031453524
However, the worst-case MEV in a 51% attack is significantly worse than the average sandwich MEV an entity can extract by using MEV boost. If the cost of censoring certain transactions, such as oracle updates, liquidations, and fraud proofs, is less than the potential profit from attacking the chain, the attack will eventually occur. These market manipulation attacks could potentially yield millions or billions in profit, and it could only take one slot to realize these gains.
There’s two aspects of security: safety and liveness. Let’s focus on where economic security is fundamental, which is to do with liveness and censorship resistance. Let’s say that I can make a billion dollars manipulating markets through selective censorship and the economic security is, $1 million. It’s a no-brainer for me to invest the $1 million to make a 51% attack and make a billion dollars of profit.
Toly disagrees about liveness, and believes it’s not a big deal. “The rest of the world runs on T+2 to T+10 settlement. It handles trillions of dollars. It has outages all the time and liveness failures, it works. But you can’t tell me that an hour outage or ten hour outage is going to stop the world and everyone’s going to be in a panic on fire”
This is not true, and it’s not something we should normalize in web3 as a ton of money has been lost due to these exact things in TradFi. The most recent example is the NYSE outage, where a technical issue caused the temporary halt in trading for dozens of stocks listed on NYSE, some of which price fell nearly 99%.
Although I’m not fully convinced by Toly’s argument, it triggered some interesting thoughts about DPoS. https://twitter.com/nickwh8te/status/1797805471061151969
Toly suggests that the economic security meme is harmless in the context of Ethereum, given its solid engineering. However, it can be dangerous for others to believe in. For instance, Terra Luna might claim they have robust economic security due to their stake, but this doesn’t matter if the underlying mechanism design is flawed.
JD predicts that some smaller chain that has much lower amounts of economic security than Solana will get 51% attacked and will be harvested for toxic MEV through, for example, the censoring of oracles.
I’ll not choose a winner as I really think that both of them exposed good arguments, but I’m going to summarize the three discussion issues.
- Timing games are real, even in fast block time networks like Solana. Underestimating this centralizing vector is dangerous and should be addressed.
- Issuance is a cost, like I said it feels more of a semantic debate here.
- I’m not sure how I feel about economic security. I definitely think it’s important, but the data cannot be taken in isolation, since there are other factors that affect the security of the chain. It might be true that a lot of PoS are overpaying for security.