Ethereum Reduced RowDAS Prototype Cuts Reconstruction Work in Simulations
Csaba Kiraly’s September 3 report shows an 11–18× reduction in estimated blob-recovery cost for a simplified RowDAS variant tested in 1,000-node simulations.
Researcher Csaba Kiraly reported on September 3 a reduced RowDAS prototype that assigns specific blob-recovery duties to high-custody nodes. The change produced an 11–18× drop in estimated network-wide reconstruction cost across 1,000-node simulations compared with the current PeerDAS baseline.
In one test case with four blobs, 10 percent supernodes, and no columns withheld, total reconstruction work fell from 48.6 CPU-seconds to 2.75 CPU-seconds. At 20 percent supernodes the figures were 91 CPU-seconds and 6.6 CPU-seconds. The figures represent accumulated processor work across the simulated network, not elapsed time.
Scope and remaining unknowns
The design keeps the existing delayed-recovery backstop on high-custody nodes and does not add new row channels. Measurements were performed only in simulated networks; no devnet results were reported. No production deployment or EIP text change has occurred. Full RowDAS row-subnet resilience is not included, and larger simulations plus real-network tests remain necessary.