Gossip Observer: New project to monitor the Lightning P2P network

The study observed a notable decrease in convergence delay among the network, dropping from approximately 500 seconds to 200 seconds for 75% propagation since similar measurements taken in 2022. This improvement is attributed to Lightning Network implementations opting for an increased number of peer-to-peer connections. Additionally, it was discovered that a considerable portion of messages received by the node originated from less than a quarter of its peers, hinting at a potential issue with the P2P connection graph or specific filtering policies within LN implementations. Specifically, channel_update messages, which constituted 60% of all messages, showed that 20% of channels had over 144 messages, indicating a higher-than-expected message volume given the proposed gossip rate limits for future updates.

In terms of data size, the unique messages collected over a single day amounted to 103.2 MB, surpassing initial expectations and highlighting the substantial bandwidth usage by nodes due to receiving duplicate messages from multiple peers. Moving forward, the researcher plans to establish permanent infrastructure to gather gossip from various geographic regions and explore broadcasting gossip messages to study their propagation further. This ongoing work will be documented in the previously mentioned gossip-observer repository.

The email also delves into considerations for improving LN's gossip mechanism, inspired by the Erlay paper and BIP suggestions. One critical aspect discussed is the adaptation of Minisketch for LN gossip, emphasizing the need for a collision-resistant mapping of gossip messages to Minisketch set elements. The proposed solution includes generating a salt for each peer connection combined with a fast hash function like SipHash to create short, efficient set elements. This approach aims to minimize computation requirements while addressing the entropy limitations associated with using transaction IDs or short channel IDs for this purpose.

Furthermore, the introduction of a gossip 1.5 message format is suggested, allowing for optional fields in gossip messages and potentially reducing implementation complexity by eliminating the need for message flooding. This shift towards set reconciliation over flooding could address issues of bandwidth waste and failed reconciliations due to divergent filtering policies across nodes.

Lastly, the researcher seeks feedback from LN implementers on several fronts, including the feasibility of adopting the per-peer state, the implications of more frequent set reconciliations, and other unforeseen challenges. The goal is to refine the proposal based on community input and draft a more concrete BIP for enhancing LN's efficiency and reliability.