Original author: IOSG Ventures

Background

Two years ago, at the beginning of the rise of modular blockchain narratives, we wrote about our views and predictions on the data availability (Data Availbility) track. As expected, the narrative of modular blockchain has prevailed and driven infrastructure innovation, enhanced network interoperability, and promoted more cooperation and integration within the ecosystem. Various Rollup as a Service (RaaS) solutions (Altlayer, Caldera, Conduit, Gelato) have begun to emerge. The following figure shows the interface of the Rollup development tool Conduit, demonstrating that deploying Rollup and selecting the DA solution has become exceptionally simple and convenient.

Source: Conduit

In the past two years, alternative DA solutions such as Celestia, EigenDA, Avail, and NearDA (Alt-DA) have made significant progress, each demonstrating unique technical advantages and market share. Meanwhile, with the launch of Ethereum EIP-4844, the introduction of blobs to replace calldata has greatly reduced the usage cost of Rollup in the Ethereum native DA layer. Today, developers and project parties face more trade-offs when choosing data availability layers. This article will track and analyze existing DA solutions, delve into their performance costs, technical characteristics, and market performance, and put forward our views and thoughts on the future development of the DA track.

1. Current situation of DA scheme

The Rollup solution on the Ethereum native DA chain mainly focuses on mainstream Layer 2 solutions that have already been updated from calldata storage to Blob-compatible, including Arbitrum, Optimism, and Base, as well as Starknet, zkSync, and Scroll, etc. Rollup uses Ethereum as the DA layer, and the data will be verified and stored by Ethereum's full nodes, benefiting from Ethereum's security, decentralization, continuity of protocol upgrades, and economic incentive mechanisms. Comprehensive L2 plays an important role in the Ethereum ecosystem, requiring the above orthodoxy brought by the native DA as a core difference. (Vitalik believes that the core of rollup is unconditional security guarantee: even if everyone is against you, you can still withdraw your assets. If data availability depends on external systems, this equivalent security cannot be achieved.)

However, publishing data to the Ethereum mainnet comes with high costs, especially before EIP-4844 (where the calldata cost is 16 gas per byte, just in December 2023, L2 spent over 15,000 ETH on DA costs). As a result, various Alt-DA off-chain solutions have emerged, such as the launched Celestia, EigenDA, and the upcoming Avail, which reduce the cost of data storage and transmission through different technical means, such as DAS, erasure coding, KZG commitments, etc.

Among them, Celestia, as a modular blockchain specifically designed for DA, has become the leading project in the DA field since its mainnet launch in October 2023. Its main target customers include projects that require modular architecture, such as cross-chain bridges, settlement layer solutions, defi projects, games, sorters, and Layer 2 solutions not limited to the Ethereum ecosystem. Its existing customers include Omnichain DEX protocol Orderly, modular L2 Manta Pacific customized for EVM-native ZK applications, Base-based L3 Hokum, and DEX Lyra and Aevo focusing on derivatives trading. As a modular design DA pioneer that is not limited to a specific ecosystem, Celestia's advantages make it the preferred choice for many emerging Layer 2 projects.

EigenDA is developed by EigenLabs, using the restaking mechanism of EigenLayer to provide an efficient, secure, and scalable DA service solution, to a certain extent inheriting the security and massive validator network of Ethereum mainnet. EigenDA focuses on providing high-performance DA solutions for the Ethereum ecosystem. As the first Active Validation Service (AVS) on Eigenlayer, EigenDA went live in April along with Eigenlay mainnet, and its existing customer base is equally diverse, including Ethereum L2 Swell, Celo, Mantle Network, and several other AVSs built on Eigenlayer, such as decentralized computing stack Versatus, Polymer, DEX protocol DODO, and Social L2 CyberConnect.

Source: EigenDA

2. The trade-offs between native DA (EIP-4844) and existing Alt-DA

2.1 Ethereum Native DA

A brief review of the development changes of the native DA solution of Ethereum. Before the Cancun upgrade, Rollup mainly used calldata as the means of data storage and transmission. Due to permanent storage and high network congestion, high costs have become the main obstacles to scalability and adoption. As the mainnet upgrade, EIP-4844 introduces a new data structure called Blob, which can accommodate large amounts of data but also increases the storage burden on nodes. Over time, storage requirements will continue to increase, eventually leading to high hardware requirements for running nodes and compromising decentralization. Therefore, Blobs will be deleted after storing for approximately 18 days (4096 epochs).

Since Blobs only require temporary storage and use a separate cost market, after the implementation of EIP-4844, the average daily cost of accessing major L2 using blob in the 60 days before and after (30 days for Scroll&Starknet) has decreased by about 99%. Among them, due to the different types of uploaded data (transaction data or state differences), the cost reduction of OP rollup's Layer 2 relative to Zk Rollup is more pronounced.

Source: Dune& Growthepie

EIP-4844 Blob capacity, storage characteristics, and pricing mechanism

Blob's capacity and storage characteristics:

• Each Block can contain up to 6 Blobs
• Each Blob can store up to 128 KB of data (even if the space is not fully utilized, the sender must pay the full Blob fee).

A new gas market for blobs, similar to EIP-1559, adjusts the base fee of blobs based on supply and demand changes.

• If the number of blobs in the block exceeds the target (currently 3), increase the basic fee for blobs.
• If the number of blobs in the block is less than the target, reduce the basic fee for blobs.

Source: IOSG Ventures

Source: Dune / Ethereum block blobs quantity 3-day moving average

L2 primarily uses the newly introduced Type 3 transactions, adding max_fee_per_blob_gas and blob_versioned_hashes fields on top of previous transactions, representing the maximum fee per blob gas that users are willing to pay and a hashed output list of kzg_to_versioned_hash, respectively.

This new pricing mechanism means that type 3 transactions still require the max_fee_per_gas and max_priority_fee_per_gas fields, and are subject to the constraints of the existing EIP-1559 market. In addition to blob space, type 3 transactions still need to pay for the EVM space they use.

Therefore, there is still competition for block space for blobs, resulting in cost uncertainty, as the blob space in each block is limited and the blob's gas fee market is dynamically adjusted based on demand.

So, as a general-purpose chain, Ethereum's shortcoming lies in the uncertainty of block space - the sudden emergence of on-chain activities such as NFT Minting and airdrop claims may cause congestion on the chain, leading to an increase in Blob pricing, making it difficult for Rollup to estimate cost basis. This will result in uncertainty in Rollup's expenditure budget, leading to unstable profitability and increasing the usage barrier for new projects that are still in their early stages, making it difficult for project parties to determine whether Ethereum DA can serve as a long-term solution. In the figure below, using blobs is about 98% cheaper than calldata for most of the time, but in the figure below, it can be seen that for a certain period, using blobs is only about 59% cheaper than using calldata.

Source: Ethernow

We calculate the cost of two blob transfers as an example:

Source: Ethernow

The figure shows a type 3 transaction of Zksync's Validator Timelock in a certain block on March 28, 2024. Let's calculate its data cost based on the blob fee, and decompose it into basic fee and priority fee according to the cost.

Assuming the price of Ethereum is $3600, the cost of using 1 Mib blob at that time was approximately:

4 × 0.018 ETH× 3600 USD/ETH = 259.2 USD

Let's take a type 3 transaction of zksync era on June 24th.

Source: Ethernow

At that time, the activity of the Mainnet decreased slightly. Let's break down the data cost:

The cost of using 1 Mib blob of data at that time was approximately: 01928374656574839201

4 × 0.0021 ETH × 3600 USD/ETH = 30.24 USD

It can be seen that there is uncertainty in the cost of using blobs to transfer data, which is still relatively high. However, for a rollup, the stability of the cost structure is one of the key considerations when choosing the DA scheme.

2.2 Celestia

As the pioneer of modular blockchain, Celestia focuses on providing DA layer and consensus layer, separating the execution layer to specifically optimize DA functionality, improve efficiency, and scalability. Compared to using the Ethereum on-chain method, Celestia as an off-chain solution L1 has many different technical characteristics, reducing the cost of data availability and providing higher flexibility and scalability. The modular design makes Celestia extremely flexible, allowing developers to freely choose the execution environment, not limited to a specific Virtual Machine (VM), enabling Celestia to support various application scenarios and meet diverse needs.

To integrate Celestia as the DA layer, Rollup needs to submit the transaction data (Data Blob) generated by the execution layer to the Celestia network instead of the original Layer 1 (Ethereum) to ensure data availability for verification and transactions. Celestia's data availability sampling (DAS) technology re-encodes block data using a two-dimensional RS erasure coding scheme, allowing light nodes to only download a small part of the block data for multi-round random sampling to verify data availability. It also allows multiple nodes to process different data parts in parallel, improving overall efficiency.

Source: Celestia.org

Another key technology introduced by Celestia in the process is the Namespace Merkle Tree (NMTs) technology, which allows different rollups to only download transaction data relevant to themselves, thereby improving data processing efficiency. NMTs not only reduce data redundancy and improve system performance, but also provide developers with more efficient ways of data processing.

In terms of security, Celestia is based on the Tendermint consensus mechanism, and validators reach consensus on Data Blob to ensure the availability and consistency of data in the network, tolerating up to one-third of validator nodes failing or behaving maliciously. By staking TIA tokens, Celestia's validators are economically incentivized to ensure honest behavior and penalize malicious or improper actions, thus ensuring the security of the network. Currently, Celestia's TVL is approximately $64.4 billion, with 100 full nodes.

Regarding scalability, the block size of Celestia can be dynamically adjusted based on the number of active lightweight nodes in the network. As more nodes join, Celestia can safely increase the block size, theoretically increasing throughput and scalability infinitely. Current data shows that its data throughput is approximately 6.67 MB/s.

Celestia Blob capacity, storage features, and pricing mechanism:

For cost comparison, we briefly discuss the performance and pricing mechanism of celestia here. When users submit data on Celestia, it is done through Blob Transaction (BlobTx), which consists of blob space fee and gas fee.

Specifically, the maximum size limit of each Blob is slightly less than 2 MiB (1,973,786 bytes), and each block can contain multiple Blobs, the specific number depending on the total size limit of the block. The current maximum block size is 64 x 64 shares (about 2 MiB), with a total of 4096 shares, one of which is reserved for PFB (PayForBlobs) transactions and the remaining 4095 shares are used for data storage. Celestia's fee market is similar to Ethereum's EIP-1559 mechanism, using a priority memory pool based on gas prices. Transactions with higher transaction fees will be processed first by validators, and the fees consist of a fixed fee per transaction and a variable fee based on the size of each Blob.

According to the comprehensive statistics of rollup data on Celestia (June 17th), for each customer integrating with Celestia, the DA cost ranges from 0.02 to 0.25 Tia/Mib, equivalent to the price of $TIA on June 17th ($7.26), and the DA cost of several major customers ranges from $0.15 to $1.82/MiB. Therefore, compared with the native DA on the Ethereum chain, Celestia provides a competitive and stable cost structure.

Source: Celenium

Source: Celenium, the gas price is stable at around 0.015 UTIA (1 uTIA = TIA × 10^(-6))

However, Celestia itself is a Layer 1 blockchain network, which requires P2P network to broadcast and achieve consensus on Data Blob. Although light nodes can use DAS to ensure data availability, the network still has high requirements for its full nodes (128 MB/s download and 12.5 MB/s upload), which brings obstacles to decentralization and future throughput improvement. In contrast, EigenDA adopts a different architecture - it does not need to reach consensus, nor does it need a P2P network.

2.3 EigenDA

As an Active Verification Service (AVS) built using EigenLayer, EigenDA utilizes the re-staking mechanism to ensure data availability by leveraging the security of Ethereum (no need to introduce a new set of validators, Ethereum validators can freely choose to join, and EigenDA's re-staking nodes are a subset of Ethereum nodes), which makes good use of the existing infrastructure. Its main workflow is that the Rollup sequencer generates Blob Data and then sends it to the Disperser (which can be operated by the rollup itself or by a third party, such as EigenLabs). The Disperser will shard the Blob Data, generate erasure coding and KZG commitments, and then publish them to EigenDA's nodes. Subsequently, EigenDA's nodes will verify the Attestation and ensure data availability. After the verification, the nodes need to store the data and send the digital signature back to the Disperser. Finally, the Disperser will collect the signatures and upload them to the EigenDA smart contract on the Ethereum mainnet for the final aggregation signature correctness verification.

The core idea is still to use technology to reduce the data storage and verification computing power required for nodes. However, EigenDA chose to implement KZG commitment verification technology consistent with the Ethereum upgrade. In addition, EigenDA does not rely on consensus protocols and P2P propagation, but uses unicast to further improve consensus speed.

And to ensure that the EigenDA node really stores the data that is available, EigenDA uses the Proof of Custody method. If there is any, anyone can submit proof to the EigenDA smart contract, which will be verified by the smart contract. If the verification is successful, the lazy validator will be Slashing.

Therefore, the solution process of EigenDA is all carried out on Ethereum, with Ethereum providing consensus guarantee, so it is not limited by the consensus protocol and the bottleneck of low throughput of the P2P network. Nodes do not need to wait for sequential ordering, and can directly and in parallel process data availability proofs, greatly improving network efficiency.

Source: Eigenlayer

EigenDA's capacity performance and cost:

The current number of EigenDA node operators is 266. Its maximum throughput target is 10 Mbps. Based on the 7-day average data, EigenDA's data throughput is 0.685 Mib/s, and the data storage and transmission cost is about 0.001 Gas/Byte. Assuming the gas cost is 10 gwei and the Ethereum price is $3600, the cost of 1 MB of data is about $0.038. The total staked TVL is 3.33 M ETH, close to $1.2 billion.

Source: EigenDA.xyz

Comprehensive comparison analysis Celestia vs. EigenDA

From a technical perspective, Celestia and EigenDA differ in several aspects. First, in terms of node load, Celestia's full node needs to handle broadcast, consensus, and validation, with a download bandwidth requirement of 128 MB/s and an upload bandwidth requirement of 12.5 MB/s, while EigenDA's nodes do not handle broadcast and consensus, with a bandwidth requirement of only 0.3 MB/s, and can use a subset of Ethereum nodes. Secondly, in terms of throughput, Celestia's maximum throughput is about 6.67 MB/s, while EigenDA aims to achieve a maximum of 10 MB/s. In terms of security, Celestia's security comes from its network value, with a stake value of about 66.5 billion US dollars and an attack cost exceeding 40 billion US dollars. EigenDA inherits some of Ethereum's security based on the re-staked asset value and the share of mainnet operators, with the current TVL approaching 1.2 billion US dollars, inheriting approximately 2% of Ethereum's security.

Overall, Celestia's competitive advantage lies in its flexible modular design and higher data throughput, making it more favored by small and medium-sized L2 and application chains. EigenDA, on the other hand, benefits from using Ethereum infrastructure, bringing orthogonality between data availability and consensus. In the future, with the development of the dual trends of modularization and application chains, Celestia may benefit from the incremental market, while EigenDA may occupy a larger share in the Ethereum center market that requires higher security.

3. Avail and NearDA

Although Celestia and EigenDA currently dominate the data availability market, the future competitive landscape may change. With the potential launch of the Avail and NearDA projects, the competitive situation in the data availability field is expected to intensify further.

Avail is a blockchain network that focuses on data availability, aiming to provide efficient transaction ordering and data storage services for EVM-compatible blockchains and Rollups. It adopts BABE and GRANDPA consensus mechanisms inherited from the Polkadot SDK. Avail uses KZG polynomial commitments as validity proofs, supports up to 1,000 validators with Nominated Proof of Stake (NPoS), and provides reliable backups through a unique light client P2P network sampling mechanism.

On the other hand, NearDA is a data availability solution launched by the NEAR Foundation, which mainly provides DA services for ETH Rollup and Ethereum developers. Its goal is to provide a cost-effective DA solution with a level of decentralization comparable to Near Protocol. It has already established strategic partnerships with major participants in the Ethereum ecosystem such as Polygon CDK, Arbitrum, Optimism, etc.

However, in the short term, the most effective way for Rollups to establish barriers is to reduce marginal costs, and adjusting the revenue and cost model according to market conditions is a better solution.

4. DA for specific scenarios

In addition to the above-mentioned general-purpose DAs for rollups, there have also emerged some early-stage DAs and DA projects tailored for specific scenarios, such as Zerogravity (0 G), a high-throughput DA solution designed specifically for AI, and Nubit, a Bitcoin-based DA solution.

4.1 Zerogravity（0 G）

The demand for data availability in AI applications is different from that in traditional blockchain applications. AI model training and operation require processing a large amount of data, including model parameters, training datasets, real-time data requests, etc. These data need to be stored and transmitted quickly and reliably to ensure the efficiency and performance of AI models. However, existing general-purpose DA solutions, such as Celestia and EigenDA, are mainly designed to meet the data availability requirements of ordinary blockchain applications, and have certain limitations in handling ultra-high throughput, low-latency massive data transmission.

ZeroGravity (0 G) aims to specifically meet the needs of AI applications through modular design and high-performance data transmission. Its modular design divides the data availability workflow into two channels: data publishing and data storage, allowing the system to scale linearly with the increase in the number of nodes. The data storage channel focuses on large data transmission, ensuring that large data can be stored and accessed almost instantly. The data publishing channel is used to ensure data availability, verified through an arbitration system based on the assumption of majority honesty. 0 G Storage is an on-chain database composed of a network of storage nodes. Storage nodes participate in the on-chain mining process through Proof of Random Access (PoRA) to ensure the availability and integrity of data. It supports the storage of various types of AI-related data, including models, training data, user requests, and real-time retrieval augmented generation (RAG) data.

Source: 0 G

0 G claims to achieve on-chain data transmission at the level of GB per second through innovative system design, far exceeding other DA solutions on the market such as Celestia and EigenDA, which have data transmission at the level of MB per second. Specifically, 0 G claims to have a data throughput of 50 to 100 GB per second, supporting scenarios such as AI model training that require large-scale data transmission.

4.2 Nubit

As the Bitcoin ecosystem gradually takes off and attracts attention, various technical routes related to Bitcoin are also emerging. With the development of these technical routes, applications such as Ordinals, Layer 2, Oracle Machine, etc., have an increasingly urgent need for efficient and secure data availability solutions. These applications need to be able to store and transmit large amounts of data quickly and reliably to ensure their normal operation and improved user experience. For example, Ordinals requires efficient data storage and transmission to support the creation and trading of digital artworks, Layer 2 solutions require high throughput and low latency to achieve better scalability, and Oracle Machine needs reliable data transmission to ensure the accuracy and timeliness of data.

Nubit is the first native Data Availability (DA) layer project in the Bitcoin ecosystem, aiming to address the limited throughput of the Bitcoin mainnet and provide infrastructure support for the long-term development of the Bitcoin ecosystem. The workflow of Nubit includes multiple steps such as data submission, validation, broadcast, storage, sampling, and consensus to ensure efficient processing and high availability of data. The data submitted by users is processed through RS encoding, verified by validator nodes using the NuBFT consensus algorithm, and generates KZG commitments. The verified data blocks are broadcasted to the entire network, and storage nodes are responsible for storing complete data blocks. Light clients verify the data availability through the Data Availability Sampling (DAS) protocol. Even in the event of network failure, nodes can recover data through full storage nodes and KZG commitments on the Bitcoin network.

Nubit is designed to provide infrastructure for Bitcoin ecological projects, and has established cooperative relationships with multiple projects such as Babylon, Merlin Chain, Polyhedra, etc. Nubit will reduce data storage costs, for example, in the case of a sharp increase in demand in the inscription market, Nubit can significantly reduce the cost of data publishing on Bitcoin Layer 2, making it more economically feasible to store and process data on Bitcoin.

5. Closing Thoughts

Analyzing the differences of projects in the DA track, we see a series of unique technologies and market positions with the widespread adoption of these DA solutions and the differences in DA layer selection among different projects, in terms of security (including data integrity, network consensus, etc.), customizability and interoperability, performance, and cost.

In the future, we believe that more App-Rollups will be launched in the market. However, although the potential market is expanding, the top effect of the DA track is obvious, and Celestia, EigenDA, etc. will occupy the major market share, leaving little opportunity for the tail. The competition is also intensifying. The current capacity is more than sufficient for Rollups. For example, after going live on the mainnet, the utilization rate of Celestia's network bandwidth has been below 0.1% for a long time, far below its maximum supported capacity of 46,080 MB per day. However, compared to Ethereum's current 15 Rollups and daily data volume of 700 MB, Celestia still has a lot of available space for activity.

Of course, it is not ruled out that there may be a demand for high-performance networks for high DA bandwidth in the future, or for example, the demand for AI projects. In addition, there are some early-stage and specific-scenario DAs, such as Bitcoin DA, which may obtain a decent market share in segmented fields. However, DA is essentially a to B business, and the income of DA project parties is closely related to the quantity and quality of ecological projects. At this stage, we believe that there is no need for excessive off-chain DA solutions in the market, unless they achieve a leap of several orders of magnitude in terms of cost and efficiency.

Overall, it can be seen that the business model of DA now has sufficient supply, but the development of the track is still evolving, and various solutions show different competitiveness in technology and market positioning. The future development will depend on the continuous innovation of technology and the dynamic changes in market demand.

References:

_US/mt-capital-research-da-sector-analysis-comparative-study-of-celestia-and-eigenda-acc0 7 ea 5694 f