Ethereum’s Fusaka-era ePBS redesign solves relay latency issues

On-chain auctions replace third-party relays

I recommend the ePBS redesign because it strips away the centralized relay model that currently controls 99% of block production. Six relays operated by five different entities handle the vast majority of blocks. The shift to on-chain auctions removes the need for middleware like MEV-Boost. In this new system, builders place collateralized bids and proposers commit to the winning payload header at the start of a slot. This change helps prevent censorship by removing the influence of the small consortium of relay operators. The transition simplifies validator work by removing coordination costs between beacon clients and relays. This upgrade moves the proposer-builder exchange into the protocol itself. Execution-layer rewards remain tied to execution payload value, including priority fees and MEV-related value, while consensus-layer rewards remain tied to validator duties like attestations and block proposals. Builders must compute proofs for up to 64 MB of rollup data, and the redesigned system facilitates this through more efficient payload handling.

Managing the free option problem

The September slot auction testnet results show that the redesign cuts proposer-builder relay latency by 40%. This reduction improves the stability of the block production pipeline. The redesign addresses the "free option" problem where builders skip revealing a payload to avoid losses during volatility. I see that this can cause the exercise rate to jump to 6.1% during market events, such as on March 4, 2025. The protocol implements a projected feedback update that raises the penalty after an exercise and relaxes it otherwise, which ensures the exercise probability stays below a target threshold while minimizing the average penalty cost for builders. You should observe the impact of these changes on validator rewards.

The protocol manages these risks using dynamic penalties. A penalty of 0.075ETH reduces exercise rates by about 75%. I find that these penalties increase entry barriers for new builders because they require more capital to buffer potential losses. This could lower proposer revenue if fewer builders participate in the auction. The mechanism achieves an average exercise probability of 0.096%, which is well below the 0.82% baseline without penalties. During high-volatility days, the mechanism drops exercise probabilities from 3.8% on August 5, 2024, to 0.6%. It also reduced rates from 2.9% to 0.5% on May 23, 2024, and from 3.1% to 0.6% on February 3, 2025. Will builders eventually find ways to bypass these dynamic penalties through off-chain agreements?

Scaling through PeerDAS and BPO forks

The Fusaka upgrade scales data availability through PeerDAS. This protocol allows nodes to verify blob data through sampling rather than downloading entire blobs. Nodes use erasure coding to reconstruct a blob from any 50% of its original pieces. This scaling path allows for higher blob targets. PeerDAS splits expanded blobs into 128 columns or cells. A validating node might only be responsible for custodying 8 or more of these 128 columns.

The network increases blob capacity through Blob Parameter Only (BPO) forks. BPO1 raises the per-block blob target and maximum from 6 and 9 to 10 and 15. BPO2 increases these numbers to 14 and 21. This capacity grows independently of major network upgrades. EIP-7883 and EIP-7823 work together to optimize the ModExp precompile. EIP-7883 raises the minimum charge from 200 to 500 gas and removes the one-third discount on general cost calculations, while EIP-7823 sets upper bounds for ModExp operations. EIP-7825 adds a transaction gas limit cap of 16,777,216 gas to prevent DoS attacks. EIP-7935 raises the default gas limit to 60 million. The RLP Execution Block Size Limit cap is 10 MiB, with a 2 MiB safety margin reserved for beacon-block framing.