Scaling Celestia through Mammoth and ZODA

Scaling through throughput increases

Celestia Mainnet Beta processes 2MB blocks with an average throughput of 0.167MB/s and 12-second block times. The Ginger v3 consensus upgrade, titled "The Doubling", reduces block times from 12s to 6s. This upgrade activates on the Arabica testnet on November 5 and the Mocha testnet later that month. The Mammoth mini testnet demonstrates 88MB blocks with 27MB/s throughput and 3s block times, providing a 160x increase in throughput compared to the current mainnet deployment which uses 2MB blocks. The Celestia team targets 1GB blocks with 83MB/s throughput. Fibre delivers 1 terabit per second of blockspace. A September 18 Mocha node update included a security fix. Ginger introduces five new Celestia Improvement Proposals: CIP-21 for blob types with verified signers, CIP-24 for versioned gas scheduler variables, CIP-26 for versioned timeouts, CIP-27 for block limits on PFBs and non-PFBs, and CIP-28 for transaction size limits. Developers can access these updates through the Celestia node ecosystem.

Light client finality and ZODA

Light clients face finality delays because they depend on the fraud proof window. To improve this, research focuses on integrating Zero-Overhead Data Availability (ZODA) into the network. ZODA allows users to sample a small number of rows and columns from a tensor-encoded matrix to verify the encoding. This method provides a way to ensure encoding correctness without needing to download the entire matrix. Such a change removes the reliance on fraud proofs and enables faster finality for light clients. You know that increasing light client participation increases reconstruction probability for the network. Light nodes perform random sampling of the Reed-Solomon encoded block data at the share level, where shares consist of 512-byte data chunks. Light clients consider a block valid once the fraud proof window for that block passes without a full node issuing a fraud proof. The ZODA protocol uses three algorithms: Encoding, Sampling, and Decoding. The Encoding algorithm incorporates randomness from a partially encoded matrix and randomizes part of the matrix before completing the tensor encoding. The Sampling algorithm enables verification by selecting small portions of the encoding. The Decoding algorithm allows anyone with sufficient rows and columns to recover the unique encoded message. The protocol uses a diagonal matrix with random 128-bit elements from GF(2^128) and the Fiat-Shamir technique. In this design, validators use the ZODA encoding algorithm to generate an extended data square containing an original data square, Q1, and parity shares, Q3, from column-wise Reed-Solomon encoding.

Competitive throughput and costs

Celestia provides a high-throughput alternative for rollups that require cheap data posting. Data shows Celestia costs 64% less than Ethereum blobs for the period analyzed. Rollups using Celestia averaged $7.31 per MB while Ethereum blobs cost $20.56 per MB. SuperBlobs from the Celestia and Conduit partnership reduce settlement costs to $0.81 per MB because bigger blobs mean rollups post fewer blobs to process the same amount of data. This reduction minimizes the frequency of settlement on Ethereum. The cost per MB for rollups using Celestia dropped significantly after they adopted SuperBlobs earlier this year.

EigenDA uses KZG commitments and provides 200ms finality. Avail utilizes ZK validity proofs and provides 20s finality. Celestia relies on a smaller, dedicated validator set compared to the Ethereum restaking pool used by EigenDA. The network uses namespaced Merkle trees to isolate rollup nodes and facilitate more efficient proofs for verification. Can ZODA effectively remove the need for fraud proofs without compromising the security of the consensus layer?