Data Availability Sampling Explained: Light Nodes & DA Layers

The Data Availability Problem

Data availability is critical for blockchain security. In rollup systems (Arbitrum, Optimism, Starknet), a sequencer posts transaction data and proofs to Ethereum. Validators verify proofs but must be able to verify the data actually exists. If a sequencer hides the data after posting proofs, users cannot withdraw funds or prove fraud. This is the data availability (DA) problem.

Why Full Nodes Cannot Scale

Full nodes must download and verify all data to confirm availability. Arbitrum posts ~1-2MB of data per minute to Ethereum. A full node verifying Arbitrum needs ~1.4 GB daily (~500 GB annually). If 100 rollups use Ethereum, nodes need 140 GB daily—impossible for most users. This limits decentralization and creates a bottleneck: only wealthy nodes can verify DA, centralizing the network.

Light Node Limitations

Light nodes (used by wallets, mobile users) do not download full blocks. Instead, they trust validators and sample block headers. But they cannot verify data availability without downloading the full data. This creates a trust assumption: light nodes must trust that validators are honest and data is available. If validators collude and hide data, light nodes cannot detect it.

How Data Availability Sampling Works

DAS solves the DA problem using erasure coding. Take a block of data (1 MB) and encode it into 64 shares such that any 32 shares can reconstruct the original block. If all 64 shares are available, the data is definitely available. If only 31 shares are available, reconstruction is impossible.

Random Sampling

A light node requests a small random sample of shares (e.g., 5 shares out of 64). The network responds with those shares. If the light node gets all 5 responses, the probability that >32 shares exist is extremely high (99.99%+). The node repeats this process dozens of times, making the statistical confidence asymptotically certain that the data is available.

Bandwidth Requirements

Full node verifying Arbitrum: ~1,400 GB/year. Light node with DAS: ~10 GB/year (100x improvement). Light nodes download only randomly sampled shares, not full blocks. The network distributes shares among peers; no single node needs to hold all shares. This allows thousands of light nodes to verify DA simultaneously without centralizing data storage.

Honest Majority Assumption

DAS assumes the network (validators, peers) is mostly honest. If >50% of nodes provide incorrect shares, the protocol can fail. But because shares are randomly distributed, a single attacker cannot control which shares specific light nodes receive. An attacker would need to control >50% of the peer network to successfully hide data—unrealistic for decentralized networks.

Celestia: Dedicated DA Layer

Celestia is a dedicated blockchain for data availability only. Rollups (Arbitrum, Optimism, others) can post transaction data to Celestia instead of Ethereum. Celestia uses DAS, allowing light nodes to verify DA without downloading full blocks. This decouples DA from settlement (Ethereum), enabling rollups to use Celestia for DA and Ethereum for settlement.

Celestia Architecture

Celestia validators receive transactions, batch them into blocks, and apply DAS encoding. Light nodes verify DAS and confirm availability. The chain's only purpose: provide DA guarantees. Celestia does not execute transactions or settle disputes—that happens on rollups or Ethereum. This simplicity makes Celestia scalable: block size can grow without increasing full node bandwidth requirements.

Cost Reduction

Posting 1 MB to Ethereum costs ~$50-200 (depending on gas prices). Posting to Celestia costs ~$0.10-1 (100-1000x cheaper). For rollups processing millions of transactions daily, this difference is massive: Arbitrum saves $10M+ annually by using Celestia instead of Ethereum for DA. Users benefit through lower transaction fees.

Rollup Flexibility

Celestia enables new rollup designs. Sovereign rollups settle to Celestia (not Ethereum), creating full autonomy. Celestia rollups can have custom VMs, execution, and settlement without Ethereum constraints. Dymension (Cosmos app-chain) uses Celestia for DA; Optimism is exploring Celestia DA. This creates a modular stack: Celestia (DA), Ethereum (settlement), custom rollups (execution).

KZG Commitments & Cryptographic Proofs

KZG (Kate-Zaverucha-Goldberg) commitments are cryptographic proofs that a sequencer has committed to specific data. Instead of posting entire transaction data, a sequencer posts a single 48-byte KZG proof. Light nodes can request data samples and verify proofs, confirming samples are legitimate parts of the committed data.

How KZG Works

A sequencer has data D (e.g., 1 MB of transactions). Using a trusted ceremony (performed for Ethereum), it generates a KZG commitment: C = commit(D). C is 48 bytes, published on-chain. To prove chunk i of D is valid, the sequencer provides a proof π. Light nodes verify: verify(C, i, D[i], π). If the proof is invalid, D[i] is fraudulent. If valid, D[i] is definitely part of the original data D.

Ethereum Proto-Danksharding (EIP-4844)

Ethereum adopted KZG for Proto-Danksharding (EIP-4844, live April 2024). Rollups can post data blobs with KZG commitments, reducing DA costs by 16x. Before EIP-4844, rollup DA costs dominated fees (~$1 per transaction). After EIP-4844, DA costs <$0.05 per transaction. This incentivizes rollups to use Ethereum over Celestia for certain use cases where Ethereum's security premium justifies slightly higher costs.

Trusted Ceremony and Security

KZG requires a trusted ceremony: randomly generated parameters are combined to create the public key. If all randomness is deleted, the ceremony is secure. Ethereum's KZG ceremony involved 150,000+ participants; the chance someone retained randomness is negligible. If randomness leaked, an attacker could create false proofs, breaking KZG security. This is the primary security assumption.

EigenDA and Ethereum Restaking

EigenDA is a DA layer built using EigenLayer, which allows Ethereum stakers to restake their ETH to secure additional services (DA, bridges, sequencers). Stakers download data and confirm availability; rollups query EigenDA stakers for DA guarantees. This leverages Ethereum's existing security without requiring a separate validator set.

Restaking Economics

Ethereum stakers earn 3-5% annually from protocol rewards. By restaking with EigenLayer, they earn additional fees from rollups (EigenDA customers). Tradeoff: restakers accept slashing risk (if they fail to provide DA, they lose stake). This economic incentive aligns staker interests with providing reliable DA. EigenDA costs rollups more than Celestia but less than Ethereum's base layer.

Adoption and Risks

EigenDA launched in 2024 with early customers (Eigenlayer partners). Risks: restaking is newer technology with unproven track record. If EigenLayer has a bug, restaked ETH could be slashed. Celestia is older and battle-tested. EigenDA offers Ethereum integration advantage (same validator set); Celestia offers independence and proven uptime. Teams evaluate risk tolerance when choosing DA layers.

Full Danksharding and DAS at Ethereum Scale

Full Danksharding (Ethereum's long-term scalability roadmap) applies DAS to Ethereum itself. Instead of a monolithic chain, Ethereum becomes a data availability layer with 64 data shards. Each shard can be verified independently via DAS, enabling Ethereum to process 100,000+ transactions per second without increasing full node bandwidth significantly.

Sharding Design

In Full Danksharding, 64 shards process data in parallel. Validators rotate among shards, sampling DA proofs across all shards. A light node samples a few blocks from each shard, verifying Ethereum's full capacity is available. This maintains Ethereum's security properties (no shard can hide data from honest light nodes) while scaling to 100,000+ TPS.

Light Client Bandwidth

Even with 64 shards at 100,000 TPS (1.6 gigabytes per second), light clients download only ~10 MB daily through DAS. This enables Ethereum light clients on mobile devices and embedded systems. Compare: Ethereum today requires light clients to download ~5 MB daily; Danksharding barely increases this despite 100x throughput increase.

Timeline

Full Danksharding is 3-5 years away (2027-2029 estimate). Intermediate steps: Proto-Danksharding (EIP-4844, done), Danksharding roadmap (blob throughput increases). Until then, rollups use Celestia, EigenDA, or Ethereum's Proto-Danksharding for DA. When Full Danksharding launches, Ethereum becomes the primary DA layer for all rollups.

FAQ