How Ethereum Rollups Actually Work

Rollups are the workhorse of Ethereum scaling. Most transactions that "happen on Ethereum" today actually happen on one. The Layer 2 overview places them in the wider family; this post opens the hood, because the mechanics explain the things users actually bump into: why withdrawals can take a week, what a sequencer outage means, and why fees collapsed in 2024.

The core move

A rollup executes transactions on its own chain, then posts two things back to Ethereum: the compressed transaction data and a claim about the resulting state. Ethereum does not re-execute anything. It stores the evidence and adjudicates disputes. Hundreds of transactions ride in one batch, splitting one L1 fee hundreds of ways, while the data on Ethereum keeps the rollup honest, because anyone can reconstruct the rollup's state from it and check the claims. That last property, data availability on L1, is what separates a real rollup from cheaper look-alikes.

The everyday flow: you send a transaction to the rollup's sequencer, which orders it, executes it, and gives you a soft confirmation in about a second. Batches go to Ethereum periodically, where they harden into finality. The sequencer is today's known weak point. Nearly every major rollup still runs a single one, which cannot steal funds (the proofs prevent that) but can censor transactions or simply go down, taking the chain's liveness with it. Decentralizing sequencers is the ecosystem's most conspicuous unfinished task.

Two ways to keep the claim honest

Optimistic: assume valid, allow challenges

Optimistic rollups post state claims with no proof attached. Instead, a dispute window of up to seven days follows each batch, during which anyone re-executing the published data can submit a fraud proof. A successful challenge rolls back the bad state and slashes the poster's bond. Security therefore rests on one honest watcher existing and on the data staying available for them to check. This is the machinery behind Arbitrum, Optimism and Base, and behind the famous week-long withdrawal: funds exiting to L1 wait out the challenge window unless a liquidity provider fronts them for a fee. EVM compatibility came almost free, which is why the optimistic camp built its lead.

ZK: prove validity up front

ZK rollups attach a validity proof (a SNARK or STARK) to every batch: compact cryptographic evidence that the state transition is correct, checked by an Ethereum contract before the batch is accepted. Invalid states cannot be posted at all, so there is no challenge window and no withdrawal delay. The trade-offs are proving cost, which is real compute though falling fast, and the years it took to fit the EVM inside a proof system, a gap that zkSync, Starknet and Polygon zkEVM have now mostly closed. The quiet consensus is that ZK is the endgame and optimistic rollups are the pragmatic incumbent. Incumbency, as usual, is proving durable.

What EIP-4844 changed

Until March 2024, rollups posted their data as ordinary calldata, competing with everyone else for gas, and that was the dominant cost of running one. EIP-4844 gave them blobs: a separate data lane, priced by its own fee market and pruned after a few weeks. That is long enough for any challenge, so security is unaffected. Rollup fees dropped roughly tenfold overnight; cent-level transactions date from that upgrade. Full danksharding extends the same lane with more capacity, covered in the Layer 1 post.

Where it is heading

Three fronts worth watching. Shared sequencing replaces single operators with decentralized sets that could also order transactions across several rollups at once. Proof aggregation amortizes many rollups' validity proofs into one L1 verification. And the two camps are converging, with optimistic systems bolting on ZK components rather than conceding. None of it changes the basic bargain that made rollups win: Ethereum's security at a hundredth of Ethereum's price, with the receipts posted where anyone can check them.