Data, Not Execution, Is Bottleneck
On-chain scaling discussions are pointing to data propagation — not execution — as the primary bottleneck, with RLNC-style coding, parallel paths, and smarter distribution suggested to speed throughput. That reframes scalability work toward network and data-layer engineering rather than raw compute alone. (x.com)
Muriel Médard — listed as co‑founder and CEO of Optimum — has been a visible proponent of treating block and blob dissemination as the primary scaling ceiling in recent talks and interviews at Consensus Toronto 2025. (coindesk.com) Random Linear Network Coding (RLNC) encodes k original data blocks into random linear combinations over a finite Galois field so peers can reconstruct full data from any sufficiently large set of coded packets, greatly reducing retransmission redundancy. (sciencedirect.com) Project literature and marketing from Optimum asserts RLNC‑based delivery plus a high‑performance pub/sub stack (mump2p) can deliver up to 20× bandwidth efficiency and roughly 2× end‑to‑end speedups versus traditional gossip/replication approaches. (okx.com) (getoptimum.xyz) Engineering precedents show the same shift in focus: Solana’s Turbine splits blocks into chunks and pushes them across parallel peers to reduce propagation latency, and Arizona State’s “Velocity” work applied rateless erasure coding to block propagation, quantifying prior practical limits (e.g., compact/XThin approaches cited with ~4–10MB tradeoffs) and demonstrating multi‑peer downloads can raise safe block sizes. (helius.dev) (blockchain.asu.edu) Recent papers on adjustable RLNC and rateless erasure codes report mathematically low probabilities of linear dependency among coded coefficients and empirically higher throughput than single‑peer full‑block transfers, providing formal support for coded, parallelized dissemination. (sciencedirect.com) (arxiv.org) System proposals now explicitly target the network and memory layers — examples include BlockSDN’s software‑defined networking integration for cross‑layer coordination and Optimum’s “decentralized RAM” concept that places coded blobs and memory access front and center as the engineering lever for throughput scaling. (arxiv.org) (prestonlabs.io)
