Data Availability: Calldata, EIP-4844, Celestia, EigenDA

What you'll learn

• The DA Problem
• The Calldata Era (Pre-Dencun)
• EIP-4844: The Blob Revolution
• Celestia: The Modular DA Layer
• EigenDA and Avail: The Alternative DA Lineup

Key terms

Data availability (DA)

The property that all transaction data needed to reconstruct L2 state is publicly accessible. Without DA, fraud proofs and ZK proofs both fail because there is nothing to verify against.

EIP-4844 (Proto-Danksharding)

The Ethereum upgrade activated March 13, 2024 (Dencun), introducing blob-carrying transactions for cheap rollup DA. Initially target 3 blobs/block, max 6; raised to target 6/max 9 by Pectra (May 2025).

Blob

A new data type in Ethereum holding ~128 KB, attached to blocks but not stored in execution-layer state. Retained for approximately 18 days then pruned. Used by rollups for batch DA.

KZG commitment

Kate-Zaverucha-Goldberg polynomial commitment scheme used to commit to blob data succinctly. The EVM has precompiles (BLOBHASH, point evaluation) for verifying blob commitments.

Celestia

A modular blockchain designed purely as a DA layer. Mainnet launched October 31, 2023. Token: TIA. Uses Data Availability Sampling (DAS) and 2D Reed-Solomon erasure coding for scalable light-client verification.

EigenDA

Data-availability AVS on EigenLayer, launched April 2024 as the first production EigenLayer AVS. Inherits Ethereum security via restaking; typically 10x+ cheaper than Ethereum blobs.

Avail

DA layer originally built by Polygon Labs, spun out independent, mainnet launched July 23, 2024. Combines KZG commitments with DAS. Token: AVAIL. Pairs with the Avail Nexus cross-chain messaging layer.

Validium

A rollup variant using ZK proofs for execution validity but off-chain DA (committee, DAC, or alternative DA layer). Cheaper than rollups but with stronger DA trust assumptions.

Data Availability Sampling (DAS)

Probabilistic light-client verification of DA. Clients sample random data points from each block; with enough samples, the probability of falsely concluding data is available drops to negligible.