Researcher Csaba Kiraly published findings on September 3 showing that an Ethereum prototype assigning blob-recovery duties among nodes reduces estimated reconstruction computing work by 11–18× across 1,000-node simulations.
Blobs carry data used by layer-2 rollups. Ethereum’s PeerDAS system allows nodes to download only part of blob data rather than the full set. High-custody nodes hold at least 64 of 128 data columns, enough to rebuild missing blob data. Supernodes hold all 128 columns.
Kiraly’s reduced design assigns particular blobs to high-custody nodes first, allowing other nodes to receive recovered data instead of rebuilding it themselves. The design uses existing column-distribution channels for sharing recovered cells.
Measured Performance
Under the PeerDAS model with four blobs and a 10% supernode share, network-wide reconstruction cost was 48.6 CPU-seconds. The reduced design cut this to 2.75 CPU-seconds under the same configuration. With a 20% supernode share, costs rose to 91 CPU-seconds under PeerDAS and 6.6 CPU-seconds under the reduced design.
Measurements on a Ryzen 9 8945HS processor showed 162 milliseconds per blob recovery. All tests were conducted in simulated, in-process networks using real cryptography. Kiraly did not report devnet or real-network results.
Relationship to RowDAS
The reduced design retains dependence on high-custody nodes and cannot provide the additional resilience of full RowDAS. Full RowDAS, specified in draft EIP-8371, would add row channels letting smaller nodes pool data and reconstruct collectively. EIP-8371 leaves blob limits unchanged.
