Scaling Ethereum Rollups with PeerDAS

Rollups depend on Ethereum to guarantee data availability so that any participant can reconstruct the state if operators fail. Without this data, users cannot verify transactions or trustlessly withdraw funds. Before the Dencun upgrade, rollups used calldata, which costs 16 gas per byte and forces transactions to compete with L1 users. This made scaling expensive. Dencun introduced blobs via EIP-4844 to provide a cheaper, temporary format. These blobs live on the consensus layer for approximately 18 days. Currently, each block contains 6 blobs with a limit of 9. Each blob holds 128KB of data. Because nodes must custody all blobs, a block with 6 blobs requires 768KB of storage per block.

The mechanism of sampling and erasure coding

PeerDAS (EIP-7594) leads toward danksharding by changing how nodes handle these blobs. The protocol uses Reed-Solomon style erasure coding to split blobs into 64 columns and extend them to 128 columns. While nodes currently download every blob in a block, PeerDAS allows a node to download only a small, randomly assigned portion of the data which provides a high probability that the entire dataset is available. Nodes subscribe to specific gossip subnets based on a unique, randomly generated ID. In the current system, a node with 8 subnets receives 1/16 of the data, but after the expansion to 128 columns, this represents 1/8 of the original data. Validators use KZG commitments to verify that the sampled data remains correct. If a node misses a piece, the extra columns allow the network to reconstruct the missing data.

The transition to PeerDAS involves updates to both execution and consensus layer clients. These changes affect blob gossiping, subnet sampling, validator custody, and cell validation logic. Node operators will subscribe to gossip subnets for assigned data slices instead of fetching entire blobs. They will also respond to sampling requests from peers while configuring custody parameters for updated software. Most importantly, the day-to-day hardware requirements remain within existing capacity for most operators.

Stake-weighted data responsibilities

Validator responsibilities change as they custody more data. For every 32 ETH staked, a node handles one extra data column. This scales with the amount of stake to maintain security. Nodes with a combined balance of at least 4,096 ETH must act as supernodes. These supernodes subscribe to all data column subnets and custody all columns to help heal data gaps.

Node Type Minimum Data Responsibility
Non-validating node 4 columns
Node with 1 validator 8 columns
Node with 57 validators (1,824 ETH) 64 columns
Supernode (121+ validators) 128 columns

Nodes staking 1,824 ETH (57 validators) handle at least 64 columns to allow reconstruction of the whole set. Nodes staking 3,872 ETH (121 validators) must handle all 128 columns.

Scaling throughput and the roadmap ahead

The 48 blob target increases throughput from ~220 to ~3,500 user operations per second. This target provides 3,300 UOPS of additional capacity for rollups. Each cell in this configuration holds 2 kB of data. Because of this, the storage requirement for a node at the 48 blob target remains 768 kB per block, matching the current requirement for 6 blobs. The network manages this via Blob-Parameter-Only (BPO) forks, which increase blob counts without a hard fork. The network uses EIP-7892 to allow client teams to coordinate small bumps without a heavy hard fork.

I find the reliance on BPO forks to step-wise increase blobs somewhat mechanical, and the shift in validator responsibility for data columns adds a layer of complexity that might frustrate some small operators. You should understand that PeerDAS does not increase the raw power of a single node, but rather its collective efficiency. Fusaka, which went live on December 3, 2025, enables this transition. The roadmap moves from Pectra to Glamsterdam in early 2026 and Hegota in late 2026. Will the increase in validator revenue sufficiently offset the increased technical complexity for solo stakers?

Newsletter