P2P networking #9 October 07, 2026

Protocol Calls & happenings
1

Agenda

– Clients updates
– Gloas Sepolia bandwidth profile
– Service discovery by Felix
– DC simulations by Sukun
– 8411 discussions (tentatively)

Meeting Time: Wednesday, October 07, 2026 at 15:30 UTC (60 minutes)

GitHub Issue

2

Meeting Summary:

The meeting covered client updates, networking optimizations, and a detailed discussion on EIP-8411 for segmented payload propagation. Daniel and jxs reported on Lodestar’s internal improvements and prototype work for attestation bundling. ashah described merged partial columns work in Prism and new peer scoring/observability features. Kamil shared X-ray-based networking analysis, noting high bandwidth and duplicate message counts in Lodestar, which Matthew said was partly addressed in recent releases. Felix presented the service discovery project in DISV5, explaining its spec status, implementation, and use cases, with Raúl questioning its necessity for CL needs. Sukun previewed decoupled consensus simulations, showing higher attestation loads and latency improvements with staggered or multi-committee publishing. The group debated EIP-8411, with Csaba and Kamil arguing for its necessity given future payload and proof sizes, and Matthew expressing concern about complexity and timing relative to H. jxs stated Lighthouse would implement EIP-8411 and attestation broadcast even if not included in H, aiming for retro-compatibility. Raúl advocated for more iterative networking upgrades, and the conversation ended with general agreement to continue the EIP-8411 discussion asynchronously.

Click to expand detailed summary

The team provided updates on various client projects and discussed network performance. Daniel reported on Lighthouse’s preparation for Sepolia and ongoing work on internal improvements and A411 prototypes. Ashah announced that partial columns work has been merged and enabled by default for Fulu and GLOAS, with Prism implementing revamps to peer scoring and connectivity that will be demonstrated in an upcoming P2P meeting. Matthew shared progress on P2P 1.3 implementation and noted that Lodestar integration would happen after Gloas goes live. Felix presented the service discovery project, explaining how it enables targeted node location through arbitrary identifiers, with the mechanism now integrated into the spec and currently under evaluation by DataHop. Raúl provided feedback on the service discovery approach, suggesting that existing peer ID and ENR mechanisms might be sufficient for consensus layer needs, particularly given current network size and scaling considerations.

Sukun presented on decoupled consensus simulations scheduled for the next day, explaining the strategy to reduce epoch length from 32 to 8 slots by splitting votes into finality votes and goldfish votes. The simulations tested three approaches on a 500-node cluster, including publishing votes at slot start, implementing a staggered publishing system with multiple committees per slot, and adjusting committee timing to reduce competition between different types of votes. Sukun noted that while the simulations showed the system works, there are still challenges with prioritizing different types of votes due to separate mesh peers, and plans to run simulations on a real subnet to gather more data.

Sukun reported running shadow simulations with 100 nodes, where 20% are super nodes and 80% are home nodes, with 1-3 validators on home nodes and more on super nodes. The team discussed scaling considerations, with Csaba explaining that while database systems can scale to millions of nodes, current simulations with 10,000 nodes haven’t encountered issues with scaling. The conversation ended with a brief update on EIP 8411, which has been merged into the EIP repository and involves chunking payloads into 64 chunks, though some teams are still reviewing the proposal.

Matthew questioned the data driving the necessity of a feature, particularly given that DevNets showed no issues with large payloads. Csaba explained that while testnets don’t provide relevant data points due to lack of home builders under stress, the feature is data-driven based on expected block sizes and bandwidths, though simulations were run with varying bandwidth scenarios including 50 megabit and 1 gigabit uplinks. The team agreed the feature seems sound and implementable, but Matthew expressed concern about the complexity adding to the timeline given the importance of other features J, K, and H.

The team discussed implementing segmented message propagation, with debate over whether to include it in the H fork or delay it to I-star. Matthew expressed concerns about implementing it in H without mainnet data to confirm the need, while Csaba and Kamil argued for implementing the basic segmentation now to create headroom for future scaling needs, particularly for post-quantum signatures. The group reached consensus to proceed with minimal segmentation and batch publishing, with Lighthouse client team committing to implement it even if not accepted for the H fork, and the discussion will continue in upcoming ACDC calls.

Next Steps:

  • ashah: Build a dashboard for Prism users to debug connectivity and demo it in the next P2P meeting.
  • Matthew: Investigate the issue with Lodestar receiving excessive copies of execution payload messages and check the version of libp2p.
  • Felix: Share the link to the service discovery implementation and spec with the meeting participants.
  • Sukun: Run the decoupled consensus simulations on a 500-node cluster and integrate a tracer for GossipSep to measure traffic.
  • Sukun: Attempt to make X-ray work with Shadow and potentially raise a PR.
  • jxs: Implement EIP-8411 and attestation broadcast in Lighthouse with retro-compatibility, even if not accepted for Hugota.

Recording Access:

3

YouTube recording available: https://youtu.be/Av6VvmP6V_k