Erlay (BIP330)

Erlay (BIP330) is a proposed bandwidth-efficient transaction-relay protocol for Bitcoin's peer-to-peer network. Today, nodes announce every unconfirmed transaction to every peer through flooding (gossip): each node tells all its connections about each new transaction ID it accepts, which means the same announcement crosses the network many times over. That redundancy is robust but expensive — transaction relay accounts for roughly half of a well-connected node's total bandwidth. Erlay's goal is to let nodes maintain many more peer connections, strengthening the network's resistance to partitioning and eclipse attacks, without bandwidth growing in proportion to connection count.

The problem with pure flooding

Under flooding, announcement traffic scales with the number of links, and most of it is waste: by the time a peer hears about a transaction from you, it has usually already heard about it from someone else. The obvious economy — connecting to fewer peers — is exactly wrong for security, since a node with few connections is easier to isolate and feed a false view of the network. The design tension is real: more connections mean more robustness and more redundant chatter. Erlay resolves it by changing how peers learn what each other knows.

Flooding plus set reconciliation

Erlay keeps a limited amount of flooding among well-connected, publicly reachable nodes — speed still matters for propagation — but routes most announcements through periodic set reconciliation. Each node maintains a small per-peer set of transactions it would have announced. At intervals, two peers reconcile: instead of listing everything, one sends a compact mathematical sketch of its set, and the other computes the symmetric difference — exactly which transactions each side is missing. The crucial property is that the cost scales with the size of the difference between the sets, not the size of the sets themselves. Two nodes that already agree on 99% of the mempool exchange only enough data to cover the 1% gap.

Minisketch and negotiation

The sketch math is implemented by Minisketch, a library built on the PinSketch construction from coding theory, tuned to decode the small set differences seen in practice in well under a millisecond. On the wire, peers advertise willingness to reconcile by exchanging a sendtxrcncl message during the connection handshake; software that does not understand the message simply ignores it, so deployment is fully backward compatible. As of recent development snapshots, Erlay remains a proposal in progress: the negotiation groundwork has landed in Bitcoin Core, but full reconciliation-based relay is not yet enabled by default — worth verifying against current release notes before citing.

Why node runners should care

The efficiency claims are worth stating carefully. The original Erlay research estimated that replacing most announcement flooding with reconciliation could cut a node's overall bandwidth substantially — on the order of forty percent at then-typical connectivity — with the advantage growing as nodes add connections, since reconciliation cost tracks mempool differences rather than link count. Those are research figures under research assumptions, not guarantees about any particular deployment. The design also deliberately keeps some flooding: rapid initial propagation among reachable nodes preserves the speed and redundancy that make the gossip layer hard to attack, with reconciliation mopping up the remainder cheaply.

For a home node on metered, throttled, or Tor-routed connectivity, transaction relay is the dominant recurring cost — Erlay directly attacks it, making it cheaper to run a node that is both frugal and highly connected. Those extra affordable connections translate into a harder-to-partition network for everyone. Erlay targets transaction announcements; block propagation was already optimized by Compact Block Relay (BIP152), and connection types like the block-relay-only connection address related goals from another angle. For privacy at the relay layer rather than efficiency, see Dandelion (BIP156).