DA is needed to create fraud proofs in optimistic rollups and escape hatches in ZK rollups. Other notes on Ethereum DA

Avail’s Background

• Founders - Anurag Arjun(ex Polygon, co founder), Prabal Baneerjee (ex Polygon )
• Avail came out of Polygon in March 2023, was building DA on top of Polkadot stack
• Launched testnet Kate in 2023, followed by the launch of Incentivised Testnet called Clash of Nodes in Nov 2023, planned mainnet first or second quarter 2024
• Polygon would offer Eigen DA, Zero Gravity and Avail etc as possible options for DA for its app chains.
• Both Celestia and Avail are targeting sovereign rollups, protocols that don't really care about execution on the base layer, but just want some ordering and data availability guarantees.

Tech

• Avail segregates data on per app basis, multiple rollups can publish their data to Avail but can query only their own data. Each roll up can choose the erasure coding ratio.
• Avail has a validator set that will sign off on published blocks but do not interpret txn data contained in the blocks.
• Avail light clients will use KZG Polynomial commitments, erasure encoding and data availability sampling (DAS) to allow verification of Data availability without downloading the entire data set. Very similar tech stack as Ethereum Danksharding.
• Rogue Validator - If a validator is byzantine, example it shares correct headers but withholds data, Avail’s Light clients would know because they sample the block. With just a few 100 or may be 1000 Light clients, entire block gets verified.
• Data is sequenced, packaged into blocks and published. KZG commitments of data matrix are generated and these commitments are then erasure encoded. A Block header should have KZG commitment and a reference to the application from where the data originated
• Ordering Guarantees - Avail provides Ordering Guarantees like Ethereum which would make it slightly slower. Ordering and DAS are tied together unlike Eigen Layer.
• Polkadot connection - Avail is built on top Polkadot Substrate and inherits their various features.
  • Consensus mechanisms : Avail’s design choice is similar to Casper and LMD GHOST used in Ethereum. LMD GHOST is Ethereum's block production engine that relies on probabilistic finality like BABE (block production engine), while Casper FFG, like GRANDPA, is a finality gadget that provides the guarantee of finality.
  • Benefits of BABE and GRANDPA -The Polkadot consensus is fault tolerant against temporary network partitions and large number of nodes crashing. BABE prioritises liveness by coordinating with validator nodes to identify new block producers. GRANDPA enables the simultaneous finalization of all blocks leading up to a specific one as soon as over 2/3 of validators attest to a chain containing that block. This consensus mechanism allows Avail to have Single slot finality in roughly about 20 seconds.
  • Nominated Proof of Stake - Avail also inherits from Polkadot that allows it to support up to 1000 validators. NPOS has speical reward distribution that mitigates the risk of stake centralization due to multi winner election method. This allows for better decentralisation of Stake and hence improving the decentralisation of the network

Competition

• Eigen DA
  • Total DA bandwidth increases linearly with the number of DA nodes in the system
  • Eigen DA does not do Ordering as they consider themselves Ethereum adjacent and they just want to increase DA throughput
  • How do you know Eigen DA nodes are storing data? Proof of Custody (Dankrad et. al), compute function (data + secret), cannot compute function if you have not stored the data and your ETH will be slashed
  • How to check if the nodes are serving the Data and not just correctly storing it? Create a Local retrieval fee market for serving. Enough non distinct parties which are non colluding. Nodes are incentivised to serve because if you do not serve someone else will serve the DA. They want large number of decentralised nodes, so Nodes have to smaller.
  • GTM is ETH L2s and L3s.
  • Cost per bit and latency would be lowest, but it needs to work with Eth.
• Celestia
  • Provides ordering and DAS. Ordering allows bridging between rollups served by same DA layer.
  • Works using fraud proofs (not KZG), slower due to obligatory challenge period in fraud proofs
  • Single slot finality but have the risk of the chain being halted when more than one third of their operators or validators are down
  • Celestia uses Tendermint so the validator set cannot go beyond a few 100 validators
  • Celestia practically invented the Modular blockchains and sovereign rollups thesis. This is one of the reasons they strong backers like Maven11 etc and the token has pumped.
• Zero Gravity TBD

Views on evaluating DA products as investments