EIP-8453: Ethereum plan to cut finality time 4x

Five Ethereum researchers have filed a draft proposal, EIP-8453 or “Decoupled Consensus”, that would give validators two separate jobs: one vote to pick the head of the chain, another to finalise blocks. Finality is the point after which a block cannot be reversed without destroying a large share of staked ETH. The draft’s initial target is a fourfold or greater cut in the time it takes to get there, and it has no activation date.

Ben Edgington opened the pull request late on Wednesday, with co-authors Francesco D’Amato, Roberto Saltini, Yann Vonlanthen and Mikhail Kalinin. “Many aspects remain TBD,” he wrote; for now the aim is an EIP number, which the draft has.

The rundown

What changed: Today one vote does two jobs under Gasper, the beacon chain’s mix of the Casper FFG finality gadget and LMD-GHOST fork choice (the rule that picks the head of the chain). EIP-8453 splits them. A small committee of about 512 validators, picked each slot in proportion to stake, casts head votes. The whole validator set sends finality attestations in their own rounds, off the epoch clock, and a block is finalised once two-thirds of active stake agrees.
Why it matters: Finality would no longer have to wait for the slot-by-slot vote rotation, so its rounds can be shortened without redesigning block production, according to the draft. Faster finality cuts the window in which blocks can be reorganised, and with it the capital bridges and exchanges tie up while they wait. The authors also pitch the split as groundwork for post-quantum signatures, which Ethereum would need if how a quantum attack on crypto signatures would work stops being theoretical. The draft does not pick a scheme.
Where it breaks: Almost everything that decides whether this is safe is still marked “TBD”: committee selection, timing and lock parameters, vote aggregation, rewards and penalties, the inactivity leak, and the full safety and liveness analysis. A 512-validator committee is cheaper to run but, the authors concede, a smaller target for denial-of-service attacks. And it needs a consensus-layer hard fork, with new block and state formats in every client.
By the numbers: An epoch is 32 slots of 12 seconds, or 6.4 minutes, and finality today takes at least two epochs, or 12.8 minutes, by our arithmetic. CoinDesk put the typical wait at about 16 minutes. A fourfold cut would mean roughly three to four minutes. The draft says the total validator rewards budget would not change, only how it is split between the two duties.

For scale, Solana switched on Alpenglow on Wednesday with a target of about 150 milliseconds. Ethereum’s nearer-term work is elsewhere; see Glamsterdam’s next testnet step. The work-in-progress spec lives on a consensus-specs branch.

Discussion: EIP-8453 on Ethereum Magicians

This article is for information only and is not investment advice.