Gossipsub
What Ethereum runs today. A node relays nothing until it holds every byte.
mump2p
Optimum's coded gossip. Nodes pass on fresh mixes before they have decoded.
Optimum Wind Tunnel · a stress test for block propagation
Ethereum validators vote four seconds into every slot. Cut a real undersea cable, black out a data centre, and count how many of them miss the block on today's Gossipsub and on Optimum's mump2p. Every shard here is really coded, and every node really decodes.
What Ethereum runs today. A node relays nothing until it holds every byte.
Optimum's coded gossip. Nodes pass on fresh mixes before they have decoded.
mump2p (formerly OptimumP2P) is Optimum's peer to peer propagation protocol. It runs inside libp2p and reuses the Gossipsub mesh, but instead of forwarding whole messages it sends coded shards made with random linear network coding (RLNC). It went live on Ethereum mainnet in October 2026 through the Optimum gateway.
RLNC splits data into k pieces and sends random linear combinations of them over a finite field, here GF(2⁸). Any k independent combinations are enough to rebuild the original. Relays can mix the combinations they hold into new ones without decoding first, so they can forward useful data as soon as the first piece arrives.
Gossipsub, used by Ethereum today, forwards a full block only after a node has received all of it, so every hop pays the full transfer time again. mump2p forwards coded shards immediately and a node needs any k independent shards, not specific ones, so slow or lost packets on one path stop mattering.
Attesters vote 4 seconds into each 12 second slot. A validator that has not received the block by then votes for the wrong head and loses rewards. Late published blocks and network trouble make this worse, which is what the cable cut scenarios show.
No. The Optimum Wind Tunnel is a fan made, open source simulation built by ecamli for the Optimum community. It follows the published OptimumP2P paper, the mump2p docs and the dev setup defaults, and it is a simulation, not a benchmark.