Skip to content
Crypto.world

The Ledger

OP Stack EIP-8288 PoC Cuts PQ Data Use From 79 Blobs to 3

A November 2 Ethereum Research post describes an OP Stack proof of concept that aggregates post-quantum signatures and proofs. In its benchmark, a 32-vCPU machine handles 170 signatures per second using an average of 2.8 to 3.3 blobs per L1 block, compared with 79 blobs when signatures travel inline.

OP Stack EIP-8288 PoC Cuts PQ Data Use From 79 Blobs to 3

A 32-vCPU machine processes 170 post-quantum signatures per second in an OP Stack proof of concept, with the resulting data averaging 2.8 to 3.3 Ethereum blobs per L1 block. Carrying those signatures inline would take 79 blobs at the same transaction rate, according to a post published November 2 on Ethereum Research.

The prototype applies EIP-8288’s signature-and-proof aggregation approach to an OP Stack chain. Rather than put large signatures and proofs into each transaction, it bundles their verification into leanVM aggregate proofs posted to Ethereum. The work is a proof of concept, not a production deployment; its measurements come from one AWS c7a.8xlarge machine running a two-second-block chain.

Why post-quantum signatures strain rollup data

Rollups move transaction execution off Ethereum, but the data needed to authenticate transactions still has to reach L1. The post argues that larger post-quantum signatures could make that data a limiting factor: in its comparison, a six-blob OP Stack channel holds about 6,600 transactions with ECDSA signatures, but just 155 with inline leanSPHINCS signatures.

The tested leanSPHINCS signature and public key total 4,968 bytes, compared with 65 bytes for an ECDSA signature. The prototype removes that material from the transaction sent to L1 and replaces it with a compact dependency identifying the statement to be proved. After compression, the post estimates that a transaction authenticated this way occupies 100 to 129 bytes, including its share of the aggregate’s statement list.

One aggregate covers multiple OP Stack blocks

The implementation adapts EIP-8288 for L2 timing. Instead of requiring an aggregate in each block header, it groups statements across multiple L2 blocks and posts the resulting proof as a new OP Stack batch type, identified as 0x02. A separate process, called op-aggregator, receives signatures and proofs from the sequencer over authenticated RPC and constructs the aggregate.

The author writes: “The prover kept up with the chain at every load, and 340 per block, the limit for a 12-block interval, gave the same result in two measurements.” In the reported 24-second aggregation interval, 340 signatures per L2 block amount to 4,080 signatures per aggregate, or 170 per second. The post reports 2.8 to 3.3 blobs per L1 block for that workload; the inline-signature comparison requires 79.

The design has a fallback if an aggregate does not arrive in time. During derivation, a block whose required proof never appears is replaced with a block containing only its deposits. That keeps the proof requirement in place, while applying it after the original block has been produced.

Lower data costs come with operating limits

The prototype charges each declared dependency as 2,048 bytes in the L1 data fee. Its author says that the fee does not cover the cost of running the prover. At full load, compute is estimated at about $0.0000027 per signature, but the post cautions that keeping a machine running can be significant for a chain with low transaction volume.

Each aggregate can hold up to 4,096 statements from as many as 64 programs. On the same machine, throughput reaches about 180 to 195 signatures per second once aggregation is included, despite signature-leaf proving alone reaching 421 per second. The post identifies the tree’s fixed-size circuits as a limit: even lower layers use circuits sized for the full 4,096-statement capacity.

Privacy proofs remain a separate test

The post also describes a devnet test using Privacy Boost, a pool that currently verifies Groth16 proofs in a contract. Replacing that route with statements included in an aggregate reduces the estimated L1-fee bytes per transfer from about 445 KB to 5.1 KB, roughly an 86-fold reduction. The proof is made by the Privacy Boost server and sent to the sequencer, rather than posted directly to L1.

The prototype does not turn the chain into a zero-knowledge rollup: it moves transaction authentication into proofs, while the chain remains optimistic and aggregates still occupy blob space. The EIP-8288 specification itself is marked draft, and its proposed L1 mechanism differs from this OP Stack implementation. Whether the prover can exceed its measured 180-to-195-signature-per-second range without raising the blob burden remains an open engineering question.

Sources

  1. Original source Ethereum Research ethresear.ch

Other pages read

  1. Ethereum Improvement Proposals eips.ethereum.org

Related stories

Explore

More articles