The future of Ethereum PeerDAS sharding and BPO expansion

PeerDAS scales blob throughput by using Reed-Solomon erasure coding to divide extended blobs into 128 columns.

This protocol uses EIP-7594 to implement data availability sampling. Regular nodes subscribe to 8 random subnets and download 1/16 of the total data. This amount equals 1/8 of the original data. I find the math sound. BPO forks change the target and maximum blob counts without requiring manual client updates. BPO1 increased the target to 10 and the maximum to 15. BPO2 increased the target to 14 and the maximum to 21. The network follows this path toward 8x capacity.

The system relies on KZG commitments to ensure data integrity. Nodes verify that received blobs correspond to the block commitment by checking polynomial evaluation points. This sampling process allows nodes to confirm data availability without downloading every full blob. Because PeerDAS divides extended blob data into 128 columns, a regular node only needs to subscribe to 8 random subnets to receive 1/16 of the total data, which equals 1/8 of the original data. The upgrade also includes EIP-7935, which raises the default gas limit to 60 million. This allows for more execution capacity. EIP-7825 limits transactions to 16,772,216 gas to prevent DoS attacks. Nodes use unique, random IDs to decide which subnets they join. The protocol distributes columns via gossip. The upgrade also includes EIP-7642, which cleans up legacy networking fields.

Stability at high throughput

The network is unstable.

MigaLabs reports that the network fails to reach target blob counts. The median blob count dropped from 6 to 4 after the first BPO update. High blob counts are rare. Slots with 16 or more blobs occur only 165 to 259 times out of over 750,000 observed slots. Miss rates climb when the network processes more blobs. At 21 blobs, the miss rate reaches 1.79%, which is more than three times the average 0.5% rate seen at lower counts. I recommend pausing BPO updates until these miss rates return to the 0.5% baseline.

The miss rate at 16 blobs is 0.77%, while it reaches 1.79% at 21 blobs. This is a significant degradation. I would not recommend further capacity increases until the network demonstrates demand at current limits. The current baseline miss rate for 0 to 15 blobs remains between 0.32% and 0.75%. The network has enough capacity for now.

Node roles and storage

Node operators manage different amounts of data depending on their role. Full nodes custody 4 columns for a moving window of 18 days. Validators custody 8 columns and add one additional column for every 32 ETH of staked balance. To be compatible with the upgrade, validators must update both their beacon node and validator client. Supernodes custody all 128 columns.

Nodes can reconstruct the entire data matrix if they acquire 50% of all columns. This reconstruction relies on the redundancy in the 128 columns. Validators must follow new fork-choice rules to accept blocks after verifying data availability. You should monitor block propagation and missed slots if you run a validator. I wonder about L2 retrieval issues.

The upgrade also includes EIP-7951, which adds native support for the secp256r1 elliptic curve. This allows for biometrics like facial recognition. EIP-7939 introduces the CLZ opcode for bit-counting. The network also includes a 10 MiB block size cap to prevent overload. These changes lead toward Glamsterdam in mid-2026. The increase in capacity helps lower fees for layer 2 rollups and improve user experience.