The latest speech at EthDenver, "The Limitations of L2," has sparked heated discussions. What is everyone debating about?

Source: @koeppelmann, @levi0214, @xcshuan, @jon_charb

Organizer: flowie, ChainCatcher

On March 4, Gnosis co-founder Martin Köppelmann delivered a speech at EthDenver titled “The Limits of L2”, in which he discussed the many limitations of L2 and another scaling solution based on zkBridge.

Martin Köppelmann's views quickly sparked widespread discussion on Twitter, with many voices of skepticism. Many Twitter users, represented by UniPassID developer @xcshuan and @dba_crypto co-founder Jon Charbonnea, believe that Martin Köppelmann's analysis of L2 is somewhat confused in logic and completely untenable; regarding the new scaling solution based on zk bridge, the skepticism was even louder, with comments filled with terms like "unreliable" and "bias-driven," and some comments mocked, "Gnosis's own chain is worth less than a testnet, processing 60,000 txs a day, while worrying about a billion users' transaction volume." In short, every core argument made by Martin Köppelmann in his speech was almost refuted one by one.

What core points were mentioned in “The Limits of L2”? What are the points of contention? ChainCatcher, in conjunction with the views and comments from Twitter users like @levi0214, @xcshuan, and @dba_crypto, has briefly organized the core points and rebuttals of “The Limits of L2”:

1. Limitations of L2

Gnosis founder Martin Köppelmann mentioned at the beginning of his speech that the initial purpose of L2 was to batch a bunch of transactions and then synchronize the results back to L1; it is a temporary space rather than a new space for storing permanent assets. In the long run, even if the initial vision is perfectly realized, L2 will still face the following fundamental issues:

Issue 1: Not suitable for state-expanding applications

Martin Köppelmann believes that L2 is only suitable for applications where the state does not expand, such as exchanges (which only need transaction results, not transaction history), but it cannot scale for state-expanding applications, citing examples of how Ethereum and L2 limit throughput.

Taking ENS as an example, if 10% (800 million people) of the global population of 8 billion want to register ENS, the entire transaction processing capacity of Ethereum would be used to handle these requests, requiring 2 years to complete. During these 2 years, Ethereum would be unable to process any other transactions.

For stocks, if all the world's stocks (45,000) use Ethereum as the settlement layer, even with L2, each stock can only conduct fewer than 30 transactions per day. This is even insufficient for the largest brokers or various transfers between L2s to settle on Ethereum.

In response to this view, @jon_charb expressed confusion, stating that the throughput issue of Ethereum L1 has been a long-standing topic; everyone knows that Ethereum L1 alone is not sufficient, and there is nothing more to say. L1 execution should continue to expand downwards (through statelessness, zkEVM, hardware/bandwidth improvements, etc.) to provide a better settlement layer, but expanding DA (data availability) is clearly the primary task.

@xcshuan directly rebutted that applications that do not expand state and do not implement mechanisms like state leasing would not encounter state explosion issues using the sidechain solution proposed by Gnosis. Moreover, Rollup does not use the state of L1; it only uses L1's DA (data availability), which actually reduces L1's state usage.

Issue 2: Gas consumption may not be significantly reduced

Martin Köppelmann believes that peak Gas on L2 is too high, sometimes exceeding $1. Even if EIP4844 is implemented, reducing Gas by 90%, there are still two issues: it is still not suitable for scenarios where Gas needs to be below 1 cent; and Gas will still rise with increased demand.

Regarding the L2 Gas consumption issue, @jon_charb's first question was, "Claiming that L2 fundamentally cannot handle transactions below 1 cent lacks evidence. In fact, no chain (L1 or L2) can guarantee transactions below 1 cent."

@jon_charb stated that EIP4844 will decouple DA (data availability) pricing from the execution layer, allowing it to scale beyond the initial parameters, and danksharding provides more DA. The initial danksharding specifications can handle thousands of rollup transactions well and can be further expanded. If the demand for Ethereum DA is very high, it cannot guarantee transactions below 1 cent.

Of course, we can also choose to eliminate the DA costs of the chain and let Ethereum provide a scalable and secure DA layer to subsidize it. DA is fundamentally a highly scalable resource, and Ethereum will ultimately provide a large amount of resources. If cheaper DA is needed for microtransactions, it may be worth considering off-chain DA, such as EigenDA, Celestia, DAC, etc.

If the execution demand of rollup is the bottleneck, then its transaction costs may far exceed the portion allocated to pay for DA, but this also applies to any chain (L1 or L2). This merely acknowledges that we need many chains, rather than that we need many L1s.

@xcshuan believes that in the absence of a silver bullet, the issue of high Gas consumption in L2 is not worth discussing. Different security assumptions are suitable for different applications, and naturally, there are different costs. Moreover, the Gas fees of Rollup are strongly related to the DA throughput capacity of L1, which is an optimizable point.

Issue 3: L2 asset withdrawal issue

Martin Köppelmann believes that small assets in L2 may not be enough to pay for withdrawal Gas fees; because the bandwidth for withdrawal is limited, if everyone wants to exit, they will be stuck.

@xcshuan countered that this viewpoint is also quite confusing. Layered networks are originally designed for each layer to perform its role; when users can meet their needs at the execution layer, why return to the settlement layer? Is it not possible to have no throughput bottleneck using sidechains + cross-chain, or can Rollup not have other bridges outside of the official bridge?

Issue 4: Some apps cannot be rollup

For this argument, Martin Köppelmann cited examples such as CirclesUBI and POA, which create a lot of states that cannot be compressed, so L2 is useless for them.

However, this argument seems to have technical knowledge flaws. @xcshuan believes that Martin Köppelmann should first clarify the difference between Data Availability and state storage before considering whether these applications cannot be Rollup.

After listing the four major issues of L2, Martin Köppelmann proposed a hypothesis: What if assets are permanently kept in L2 and treated as a permanent space? Would that work?

In response to this hypothesis, Martin Köppelmann raised new questions about L2.

1. L2 sequencers are very centralized

For example, while L2 cannot take your money, they have immense power—deciding whether to accept your transaction, how much Gas to charge you, and the order of transactions. Martin also took a jab at Coinbase, saying that if you want to operate an exchange on Base, it can easily place its transactions ahead of yours.

At the same time, centralized sequencers are very likely to be censored, or even enforce KYC (only accepting transactions from KYC addresses). Martin emphasized that this is very likely from the current regulatory perspective.

Regarding the centralization issue, @jon_charb and @xcshuan both unanimously rebutted that while sequencers are currently centralized, they can be designed to be decentralized in future upgrades.

@jon_charb mentioned that sequencers can not only be decentralized but can also add upgrade delays so that users can exit before upgrades they disagree with. Meanwhile, @xcshuan questioned whether the sidechain solution Martin mentioned—using dozens of nodes with BFT (Byzantine Fault Tolerance consensus algorithm)—is really decentralized, and whether Rollup can connect to BFT's PoS?

2. If L2 issues native assets that do not exist on L1, what is the significance of L2?

Martin Köppelmann explained that L2 is secure because it inherits security from L1, but if you don't want L1, why do you need L2?

@xcshuan stated that this question seems interesting but appears to lack clarity. There is a conceptual confusion; the purpose of Rollup is to achieve certain scalability and a new execution layer while using L1's security.

3. Rigidity issue

Martin Köppelmann mentioned that Ethereum L1 itself is still evolving and will undergo many modifications in the next 5-10 years, which poses many challenges for L2.

Martin Köppelmann provided an example to illustrate: the Snapshot we use for voting is conducted on L2, and then the state is synchronized back to L1, where they perform Merkle Proofs. However, Ethereum L1 plans to switch from Merkle Trees to Verkle Trees in the next year or two, which will render the current version of Snapshot unusable. Therefore, L2 may need some kind of "upgrade" mechanism, but this would contradict its trustless goal.

In response, @xcshuan pointed out that the settlement layer should be designed to be sufficiently abstract, and L2 should not use assumptions below the abstraction layer; the abstraction layer should always be backward compatible.

2. New Scaling Solution: Ethereumverse

Martin Köppelmann believes that while L2 has its merits, we need more L1s, thus proposing a model similar to Cosmos's IBC, planning to create another sidechain that runs the same content as Ethereum and connects through a trustless ZK-bridge, forming an Ethereum universe (Ethereumverse).

Using zk to run a light node client of one chain on another chain, and then verifying without trust, is far safer than traditional bridges.

The alternative new scaling solution and concept proposed by Martin—Ethereumverse—has indeed sparked some anticipation among Twitter users.

However, there are also widespread doubts: First, while L2 currently has many issues, it cannot be denied that L2 provides better scalability/security/interoperability for the vast majority of use cases.

Second, the problems of L2 cannot be solved by launching a bunch of new L1 execution layers. @jon_charb believes that the cross-chain security between L1s is low, and it is best to solve it by expanding Ethereum's DA/execution + rollups. @xcshuan also believes that zkBridge has not broken through the security assumptions of the three types of cross-chain solutions: bidirectional light nodes, relays, and hash time locks. Although using succinct proofs can compress computations that are otherwise unfeasible or very expensive to verify on-chain, it is still either a bidirectional light node or a relay. Without succinct proofs, it is still possible to design cross-chain solutions with the same security model.

In summary, sidechains, channel networks, cross-chains, and various modular networks for settlement/execution/DA are sometimes viable technical options. However, one should not disregard technical logic in favor of promoting a particular path and selectively describe the situation.