Somnia decouples data production from agreement. Each validator continuously appends to its own “data chain” with no local consensus, while a separate BFT consensus chain periodically snapshots all heads and imposes a single, deterministic global order. This enables streaming-style propagation with heavy compression and signature aggregation, versus most chains that propagate one leader’s block per slot via gossip or tree-fanout.
How Somnia’s pipeline works
Per‑validator data chains: Every validator publishes its own byte‑blob blocks at high rate without waiting for votes; these streams can advance nearly continuously and in parallel across the network. Inclusion is immediate at the data layer.
Consensus chain for ordering: A modified PBFT proof‑of‑stake chain collects the latest head of each data chain at tight intervals and finalizes a block that semantically includes all transactions up to those heads. One pseudorandom deterministic merge produces a single ordered sequence.
Decoupling benefits: By separating production and ordering, Somnia removes the leader bottleneck, turns propagation into steady streams, and unlocks cross‑block streaming compression to cut bytes dramatically.
Data propagation versus others
Streaming compression: Continuous per‑validator streams allow compression across time, with materials claiming up to ~20× better node‑to‑node data movement efficiency than typical blockchains during tests. This reduces bandwidth pressure at high TPS.
Signature aggregation: Batches aggregate many signatures (e.g., BLS) into compact proofs, shrinking payloads and verification costs compared to verifying each transaction signature independently.
Contrast with Solana: Solana disseminates leader blocks using Turbine, a tree‑based fanout that shards block data into “shreds” and pushes via UDP/QUIC to neighborhoods to minimize redundant sends. It’s still a single global block per slot rather than per‑validator streams merged later.
Contrast with Ethereum‑style gossip: Ethereum propagates whole blocks via peer gossip over TCP/DevP2P, optimized for longer slots and smaller blocks; Somnia instead pushes many small, continuous updates per validator and compresses them across time.
Consensus semantics and latency
Fast BFT ticks: The ordering layer runs short rounds, finalizing a snapshot of all chain heads every tick; because the data was already propagated in streams, most bytes are “in place,” so consensus confirms rather than ferries bulk data. This supports sub‑second finality.
Security model: Although data chains have no safety rules themselves, the consensus chain’s inclusion of a single head per validator prevents double‑publishing from becoming canonical and enforces one coherent history.
Why it’s different in practice
Throughput scaling: Intake scales with the number of validators publishing in parallel; ordering remains lightweight because it references heads and merges deterministically, rather than coordinating a monolithic block’s contents every slot.
Network efficiency: Streaming compression plus signature aggregation means less bandwidth per transaction as load rises, improving fee stability during spikes compared to naive gossip.
Bottom line: Most chains either propagate one leader’s block per slot with clever fanout or gossip, then finalize; Somnia pushes parallel validator streams continuously and runs a tight BFT to lock in an interleaved global order. That architectural split, paired with compression and aggregation, is what differentiates its consensus and propagation from peers.