Author: PSE Trading Analyst, @cryptohawk
1.1 An essential part of the modular blockchain
The historical mainstream blockchain architecture is a non-hierarchical structure, that is, the four core functions of execution/settlement/consensus/data availability are performed by the same batch of nodes. On the contrary, the hierarchical structure of the blockchain, the node only needs to focus on a part of the four core functions of execution /settlement/consensus/data availability, thereby reducing the node hardware threshold and achieving network expansion.
The definition of the four core functional modules of the blockchain:
At the moment when the Ethereum Rollups ecology with Ethereum as the core is in full swing, the gas fee of L2 can save nearly 90% compared with that of L1, but it is still not low enough. There is still a distance from the goal of connecting billions of users envisioned in the future.
According to the speech of the founder of Avail at the ETH Community Conference in July 23, nearly 70% of the cost of Rollups currently lies in the uploading of tx data & proof data on Ethereum L1. The next step for modular blockchains is almost foreseeable, with ETH L1 and numerous dedicated DA layers competing at the level of data availability to dramatically lower the barrier to entry for new rollups, further strengthening the blockchain scalability and reducing interaction costs without sacrificing security and decentralization.
1.2 The latest development of DA layer
1.2.1 DA layer technology path
Regarding how to ensure the availability of data, the DA layer has adopted many technological innovations, and some technical directions have been agreed by almost all DA layers, particularly about how to ensure that complete data can be obtained/restored:
(1) Erasure code
In order to prevent the DA node from losing data fragments, the erasure code technology expands the original data from N elements to M elements (M > N), as long as any N unique elements are obtained from the M elements of the extended data, the complete extended data.
The DA layer uses tx/blob in the block as the smallest element, EigenDA & Espreeso adopts the one-dimensional Reed-Solomon coding scheme, and Celestia & ETH Darksharding adopts the two-dimensional Reed-Solomon coding scheme.
(2) Data availability sampling
The data availability sampling mechanism is based on erasure codes, that is, nodes do not need to download complete block data, and a certain number of data fragments are randomly sampled from block builders by sufficient nodes (even light nodes) to ensure that full blocks can be recovered even in the worst case.
Of course, in other technical directions, such as how to prove that the original data has been correctly coded and expanded, there are differences in the schemes adopted by different DA layers:
(1) Fraud Proof Mode
Representative project: Celestia
Through the light nodes sampling enough unique data fragments and broadcasting them to the full nodes, the honest full nodes can execute the encoding to restore the complete block, recalculate the Data Merkle Root and compare it with the Root issued by the block builder for verification. If the verification fails, that is, it proves that the original data has not been correctly encoded and expanded, the full node will broadcast the fraud proof to other light nodes & full nodes.
Advantages: The technical barriers to the implementation of the game theory incentive mechanism are smaller;
Disadvantage: Minimal honesty assumption needs to be met.
(2) KZG Commitment Model
Representative projects: EigenDA, Espresso, Avail, ETH Darksharding
KZG commitment is a polynomial commitment proof. According to the tx data specification of the DA layer, all original data and extended data are mapped to the X, Y grid, as shown in the figure below with 8 elements (d0, x0),(d1, x1 )...(e0,x4),(e1,x5)..., and then find the minimum degree polynomial that runs through them through Lagrange interpolation. Prover needs to make a commitment to this polynomial f(x) with a secret trusted setting C(f).
Subsequently, Prover will generate a fixed 48-byte proof Π for the group elements. With C(f), the verifier can verify whether y=f(x) is true for each element, provided that all the element points are all on the same polynomial, it can prove that the original data has been correctly coded and expanded.
Advantages: Fast verification;
Disadvantages: Trusted settings are required in advance & not resistant to quantum computing.
(3) DA committee multi-signature mode (Data Availability Committee)
Representative project: Arbitrum Nova
In this mode, the blockchain relies on an external DA committee to store tx data and promises to provide data according to the needs of users. The DA commitment means that the committee members sign the Hash & DA expiration time of a certain tx data block with a BLS signature that meets a certain number of thresholds.
Advantages: extremely low cost;
Disadvantages: Data availability also depends on the matching honesty incentive model + malicious punishment model + DAO governance model, so the reliability is lower than fraud proof & KZG commitment model. It is suitable for low-value tx data storage of non-financial applications.
1.2.2 Module division scheme
There are also many design schemes for how different modules of the blockchain are divided between specific projects. The Six mainstream design frameworks, Celestium, Celestia Sovereign Rollup, Eigen Rollup, Espresso Rollup, and Ethereum Rollup, are as follows.
Throw out a few core points:
(1) The unified settlement layer enables many Rollups to enjoy cross-chain security & aggregated liquidity.
Compared with the cross-chain between L1 through the relay trust layer, rollups with the unified settlement layer can share the global state between each other in real time at the settlement layer, so the security of token & information transfer between blockchains is higher.
The author lists two kinds of third-party cross-chain bridge solutions:
- Through the official cross-chain bridge contract & SDK under Rollups, the cross-chain without greater trust assumption is realized;
- Achieving faster and lower-cost cross-chain through the third-party liquidity pool.
(2) The tx sequencing right should not be assumed by the DA layer.
Recently, Celestia researcher NashQ proposed a variety of Rollup variant modules, in which the tx sequencing right is mainly assigned to the aggregator/DA layer. The author believes that more and more attention has been paid to the discussion on the democratized distribution of mev. The mechanism represented by PBS can reasonably distribute the value of mev among arbitrageurs & nodes/sequencers, and will be adopted by the top Rollups with a high probability. The design of the consensus mechanism and network architecture of the DA layer should focus on the guarantee of data availability. If an additional mev allocation mechanism related to tx sequencing is added, it may pose unnecessary challenges to the technical requirements of the network architecture.
(3) Ethereum will still be the best option for most Rollup consensus layers & settlement layers in the next ten years.
Under the modular blockchain framework, most blockchain users (even industry professionals) don't care much about the security & block finality provided by the consensus layer, but the author believes that the consensus layer is the core of the blockchain module. Even in 2023, there was an abnormal block reorganization event of the Polygon POS chain, greatly extending the block confirmation time of the centralized exchanges & cross-chain protocol for the Polygon POS chain, which had a negative and far-reaching impact. Therefore, the author predicts that Ethereum will be unshakable as the leader of the consensus layer of the smart contract public chain (maybe over 90% market share) in the next ten years, which could be the best option for the consensus layer of Rollups. Besides, Rollups had better adopt the same blockchain as the settlement layer and the consensus layer.
1.3 Introduction to Mainstream DA Layer Projects
1.3.1 Celestia
As the first solution to provide a DA layer, Celestia's network architecture is divided into a consensus layer and a data availability layer.
(1) Consensus layer: Celestia borrows heavily from the architecture of Cosmos, and builds a POS chain named Celestia APP as the consensus layer, under which Celestia-core adopts the modified version of Tendermint as the consensus algorithm, and the nodes obey Tendermint p2p network rules, and connect to the application layer (that is, state machine) through ABCI++ to execute PoS logic and perform governance.
(2) Data Availability Layer: Celestia uses Data Availability Sampling (DAS) technology to allow light nodes to generate security attributes nearly equal to full nodes by only downloading block headers containing block data Merkle root without downloading complete blocks. Specifically, in each round of DAS, the Celestia light node will sample the 2k × 2k data blocks encoded with erasure codes for each block. Each light node randomly selects a set of coordinates in the expansion matrix, and queries the full node for the data fragments and the corresponding Merkle proof at these coordinates.
Assuming that the full node hides the tx when broadcasting a block containing 1000 txs, under the assumption that the light node is required to pass the sampling test to be available for all block data (that is, no error is detected/not too many data fragments lost to restore the complete block) with 99.9999% confidence probability, if simple sampling of 1000 original data blocks & a malicious full node hides 1 tx, it will take about 13,800 times of samples to achieve, it is better to directly download the complete block; and if 4 million extended data fragments are sampled & the malicious full node hides more than 1 million data fragments, only 48 times of samples are needed. The efficiency difference is about 288 times.
What DAS can achieve:
-
A small amount of sampling can find out whether the block broadcast by the full node hides more than 25% of the block data;
-
Sampling to obtain 75% of the data can ensure that the complete block data can be recovered.
What DAS cannot achieve:
-
If the block producer hides more than 25% of the data, it may not be possible to restore the complete block data;
-
If there are not enough light nodes to sample, it may not be possible to sample enough non-duplicated data blocks to reconstruct the entire block.
1.3.2 EigenDA
As the first AVS network officially developed by EigenLayer, EigenDA is positioned at the DA layer with the security subset of Ethereum, and mainly promotes the sovereign blockchain solution in which the settlement layer is concurrently held by the execution layer.
The founder of EigenLayer, Sreeram Kannan, has conducted innovative research on Coded Merkle Tree, Scalable Data Availability Oracle, DispersedLedger and other technologies on DA, and currently adopts block data double-redundancy one-dimensional erasure code + KZG commitment + Authenticated Coded Dispersal (ACeD), the technical framework for single-node storage of 1/n data blocks (the number of network nodes n). EigenDA is expected to greatly exceed the final DA solution Danksharding of ETH in terms of DA efficiency and node bandwidth.
1.3.3 Espresso
The Espresso Sequencer network chooses to separate the DA layer and the consensus layer in a modular manner under the same node set. The DA layer is responsible for sequencing tx + ensuring data availability, and the consensus layer is only responsible for reaching an agreement on the short commitment of the data set. In addition, the DA layer and the consensus layer will also rent/share the security of ETH through a re-staking layer such as EigenLayer.
Advantage:
(1) Flexibility: Under optimistic conditions, the CDN & small DA committee can greatly improve the data transmission capacity & block confirmation speed of the network. Under pessimistic conditions, the network can also switch to the P2P protocol & DA base layer in time to ensure security;
Shortcoming:
(1) Architecture redundancy: the Espresso Sequencer consensus layer does not need to be separated from the DA layer;
(2) The network security is almost equal to the amount of ETH pledged by EigenLayer in the Espresso Sequencer network, and there is a risk that EigenLayer's resources tend to be tilted towards EigenDA when Espresso competes with EigenDA on the same track;
(3) MEV capture capability & transaction sequencing/censorship rights are completely concentrated in Tiramisu, that is, the Espresso DA layer, which needs to be decentralized by solutions such as PBS in the future.
1.3.4 ETH Proto-Darksharding
In the future roadmap of Ethereum shown by Vitalik on 22.11.5, it clearly shows that after The Merge: POW to POS stage, the key goal of Ethereum at next stage is to further improve the transaction performance for Rollups through EIP4844. Currently the entire Ethereum is positioned as DA & consensus & settlement layer, and only the execution layer is appointed to Rollups.
EIP4844 is expected to be launched in the Cancun upgrade at the end of the year. This EIP introduces a new transaction type, blob-carrying transaction. The tx data uploaded by Rollup can be stored non-permanently on ETH Layer 1 in the form of blob. The size of a single blob is 128KB. Ideally, each block contains 8 blobs, with a size of about 1MB, and a maximum of 16 blobs, with a size of about 2MB. Compared with the current average block size of ETH of 90KB, a special expansion will be realized. In order to prevent the explosion of the storage status of ETH nodes, it is planned to automatically delete blobs beyond a period of time (the specific time window has not been determined, it may be 2 weeks or 1 month), so blobs can be regarded as a kind of cache.
Although the future finality envisaged by Vitalik only has the state root stored on the Ethereum chain, and detailed transaction data is stored on the dedicated DA layer, the short-term compromise solution EIP4844 will point to the direct business competition of the ETH chain and the dedicated DA layer. In addition to "lower data storage cost", whether the dedicated DA layer can explore a broader business model and build a better DAPP ecosystem will be the key to success.
1.4 Conclusion
In the past round of cryptocurrency cycles, the entire data storage track did not have an advantage in the accumulation of capital bubbles and attractiveness to developers, because users were not sensitive to the risks of centralized data storage & hosting, and the need for decentralized storage is temporarily falsified. The DA layer, as an essential module in the modular blockchain, is featured as the storage layer of the most valuable transaction data, and ensures the availability of data (public access without access & anti-censorship) & Integrity & Correctness & Privacy at a lower cost, which will be a more attractive narrative.
In the short-to-medium term, the DA layers will experience fierce competition.
(1) After the Cancun upgrade, Ethereum Rollup will benefit from the storage costs reduction in blob data brought by EIP4844, and may continue to maintain the market competitiveness of ETH L1 in a perspective of DA module;
(2) DA layer solutions (such as Celestium) that use ETH L1 as the settlement layer can bring better "Lego building block" interoperability between execution layers by enabling the execution layer to share cross-chain security/liquidity, contributing to a virtuous circle of ecological development;
(3) DA layer solutions (such as EigenDA, Espresso) relying on the EigenLayer re-staking protocol, can not only reduce the cost of tx data storage, but also share part of the security of ETH L1;
(4) The shared sequencing scheme with a good incentive distribution mechanism (such as PBS) will be adopted by the top Rollups. The tx sequencing/censorship right should not be controlled by the DA layers, since the DA layers should concentrate on the task of making data available.
References:
2.'Unpacking Celestia'， from Analyst DAO,
3.'Pay Attention To Celestia', by Can Gurel, from Delphi Digital
https://members.delphidigital.io/reports/pay-attention-to-celestia/
4.'State of Modular Blockchains', by Roy Lu,
https://roydeploy.substack.com/p/state-of-modular-blockchains
5.'Fraud and Data Availability Proofs: Maximising Light Client Security and Scaling Blockchains with Dishonest Majorities', by Mustafa Al-Bassam, Alberto Sonnino, and Vitalik Buterin
https://arxiv.org/abs/1809.09044
6.Data availability sampling and danksharding: An overview and a proposal for improvements，by Valeria Nikolaenko and Dan Boneh
7.Exploring MEV on EigenLayer，by Walt Smith
https://www.galaxy.com/galaxyperspectives/exploring-mev-on-eigenlayer/
8.Hack Summit 2023 How to build new VMs and rollups using eigenDA
https://www.youtube.com/watch?v=B2qBnLmPupQ
9.EigenLayer: The Restaking Collective，by EigenLayer Team
10.Don't overload Ethereum's consensus
https://vitalik.ca/general/2023/05/21/dont_overload.html
11.https://hackmd.io/@EspressoSystems/HotShot-and-Tiramisu
12.https://github.com/EspressoSystems/HotShot/blob/main/docs/espresso-sequencer-paper.pdf
13. HotStuff Consensus algorithm details
https://blog.csdn.net/ganzr/article/details/110879463
14. The Hitchhiker's Guide to Ethereum, by Jon Charbonneau
https://members.delphidigital.io/reports/the-hitchhikers-guide-to-ethereum/#key-takeaways