Skip to content

Guide · Protocol

Data availability: blobs, DAS and the DA-layer debate

Why data availability is distinct from correctness, how Ethereum's blobs provide it today, what data availability sampling (DAS) changes, and the tradeoff of external DA layers. Not financial advice.

By Etherelysium Editorial TeamUpdated 18 November 20267 min read

Data availability: blobs, DAS and the DA-layer debate
Share X LinkedIn

A rollup’s security claim only means something if the data behind its batches is actually available for someone to check. If a sequencer could post a state root without publishing the transactions that supposedly produced it, nobody could ever generate a fraud proof or verify a validity proof against anything real — the proof system would be checking a claim nobody can independently audit. Data availability is the property that closes this gap: the guarantee that the data behind a claimed state update is actually published somewhere anyone can retrieve it.

Why availability is a distinct property from correctness

Correctness and availability are two separate questions. A validity proof can mathematically guarantee that a state transition is correct relative to some claimed input — but if that input was never actually published anywhere, nobody can confirm the proof was checked against the real transaction data rather than something else entirely. Fraud proofs depend on availability even more directly: a challenger cannot construct a fraud proof against data they cannot access in the first place. This is why Ethereum built a dedicated mechanism for availability specifically, rather than treating it as something that falls out automatically from having a proof system.

How Ethereum provides DA today: blobs

Since EIP-4844, rollups post their batch data into blobs — large, temporary data packages carried alongside L1 blocks, covered in depth in the companion guide on blob fees. Every full node that downloads a block also downloads its blobs, which is a direct but bandwidth-heavy way to guarantee availability: the whole network simply holds a full copy of the data for a retention window (roughly 18 days) before pruning it.

The roadmap: data availability sampling (DAS)

Requiring every node to download every byte of every blob does not scale indefinitely — it caps how much data the network can make available without pushing hardware requirements for running a node out of reach of ordinary participants. Data availability sampling is the proposed fix: rather than every node downloading all the data, each node downloads and verifies only a small random sample of it. Because of how the data is encoded (using erasure coding, which allows the full original data to be reconstructed from a sufficient subset of samples), if enough different nodes across the network each hold their own small sample and those samples check out, the network can be statistically confident the full data is genuinely available, without any single node needing to store all of it. This lets the total amount of data the network can support grow well beyond what “every node downloads everything” could ever sustain, while keeping the hardware bar for running a node low enough for broad participation. EIP-4844 was deliberately built to be format-compatible with this future design, specifically so today’s blobs would not need to be restructured when sampling-based scaling eventually arrives.

The debate: on-chain DA vs. off-chain DA layers

Not every rollup posts its data to Ethereum at all. Some use separate, external data availability layers instead — specialized networks built specifically to store and attest to data availability, independent of Ethereum’s own consensus. The appeal is straightforward: it can be cheaper and sometimes faster than competing for blob space on L1. The tradeoff is just as straightforward: a rollup’s security becomes partly dependent on that external layer’s own honesty and uptime assumptions, rather than resting entirely on Ethereum’s validator set, which is a meaningfully different and additional trust assumption layered underneath whatever proof system the rollup itself uses.

A rollup using an external DA layer is sometimes described informally as a “validium” rather than a full rollup, precisely because of this distinction — the execution and proof side may look identical to a rollup posting to Ethereum, but the availability guarantee underneath it does not. Whether that tradeoff is reasonable depends entirely on how robust and how independently verified the specific external DA layer actually is, which is a project-specific question rather than a generic answer this guide can settle.

Why this matters for reading an L2’s security

Two rollups can use the exact same proof system and still differ meaningfully on this one axis: where their data actually lives, and under what guarantee. A rollup posting to Ethereum blobs inherits Ethereum’s own validator-set security for availability; a rollup relying on a separate DA layer inherits that layer’s security instead, whatever that happens to be. This is a distinct, separately checkable fact from the rollup’s stage or proof type, covered in the companion guide on L2 security models.

A simple mental model

Think of a rollup’s batch as a claim with receipts attached. A validity or fraud proof checks the arithmetic of the claim; data availability is the guarantee that the receipts themselves are actually sitting somewhere auditable, rather than the claimant simply asserting they exist. A proof system without availability is an auditor checking a company’s math homework without being allowed to see the company’s actual books — the arithmetic can be flawless and still meaningless if nobody can confirm what it was computed from.

Retention is a deliberate tradeoff, not an oversight

Ethereum does not keep blob data forever, as covered in the companion blob-fees guide, because permanent on-chain storage for data that only needs to be checked once would make running a full node progressively more expensive over time with no corresponding security benefit after the checking window has passed. The roughly 18-day window is a considered balance between giving the ecosystem enough time to verify and archive what it needs, and keeping the base protocol’s own storage burden bounded and predictable for the long run.

Long-term historical access to blob data, once the retention window passes, depends on third-party archives choosing to keep and serve it — a separate, additional layer of reliance worth knowing about for anyone who needs to verify old, rather than current, rollup activity.

Checking which specific mechanism — Ethereum blobs or a named external layer — backs a given rollup takes only a quick look at its own documentation, and is one of the more reliable single facts available about how its proof system is actually anchored.

It is a smaller time investment than it sounds, and it tends to be far more informative than the proof-type label alone.

And it is a check worth repeating periodically, not just once, since a rollup’s DA arrangement can change as it evolves.

Data availability is, in the end, the quiet precondition everything else in this guide series depends on — without it, none of the proof systems or finality guarantees described elsewhere on this site would mean anything at all.

Risk note

Using a separate, off-chain data availability layer is a real and common design choice, not an automatic red flag — but it is a different, additional trust assumption from relying on Ethereum’s own validator set, and it is worth knowing which one a specific rollup has actually made before treating its security as equivalent to a rollup that posts to Ethereum directly.

Not financial advice. Nothing on this page is a recommendation to buy, sell, stake, bridge, or avoid any asset or protocol.

FAQ

Is data availability the same thing as a validity proof?

No. A validity proof checks that a state transition is mathematically correct relative to some claimed input. Data availability is the separate guarantee that the actual input data was published somewhere anyone can retrieve and check it against.

What does data availability sampling actually change?

Instead of every node downloading every byte of every blob, each node checks only a small random sample. Combined with erasure coding, enough independent samples checking out gives the network statistical confidence the full data is available, without requiring any single node to store all of it — letting total data capacity scale well beyond what full-download requirements could sustain.

What is a validium?

An informal term for a rollup-like system that relies on an external, off-chain data availability layer instead of posting its data to Ethereum directly. Its execution and proof mechanics can otherwise resemble a standard rollup.

Is using an external DA layer less secure?

It adds a different, additional trust assumption — confidence in that specific external layer’s own availability guarantees — on top of whatever the rollup’s proof system already requires. Whether that tradeoff is acceptable depends on how robust that particular external layer actually is.

Frequently asked questions

Is data availability the same thing as a validity proof?
No. A validity proof checks that a state transition is mathematically correct relative to some claimed input. Data availability is the separate guarantee that the actual input data was published somewhere anyone can retrieve and check it against.
What does data availability sampling actually change?
Instead of every node downloading every byte of every blob, each node checks only a small random sample. Combined with erasure coding, enough independent samples checking out gives the network statistical confidence the full data is available, without requiring any single node to store all of it -- letting total data capacity scale well beyond what full-download requirements could sustain.
What is a validium?
An informal term for a rollup-like system that relies on an external, off-chain data availability layer instead of posting its data to Ethereum directly. Its execution and proof mechanics can otherwise resemble a standard rollup.
Is using an external DA layer less secure?
It adds a different, additional trust assumption -- confidence in that specific external layer's own availability guarantees -- on top of whatever the rollup's proof system already requires. Whether that tradeoff is acceptable depends on how robust that particular external layer actually is.

Educational content, human-edited and possibly AI-assisted. It describes how systems work and is not financial, legal or tax advice. Report an error via our corrections page.

More in Protocol

Related articles

See all