Thanks Emmanuel Nalepa for the review and contribution.
TL;DR
This doc aims to analyze PeerDAS sampling spec and propose a change to the spec to mitigate the underlying issue.
Background
EIP-7594: PeerDAS - Peer Data Availability Sampling proposes an intermediate step towards full Danksharding, its design separates network interaction between nodes into two phaes: distribution and sampling.
During distribution phase, blobs were extended horizontally (1D PeerDAS), and then divided into NUMBER_OF_COLUMS (currently 128) sections and distributed across DATA_COLUMN_SIDECAR_SUBNET_COUNT (currently 32) subnets. Each node is required to custody at least CUSTODY_REQUIREMENT (currently 1) subnets (4 columns).