P2P networking #8 September 09, 2026

Protocol Calls & happenings
1

Agenda

Meeting Time: Wednesday, September 09, 2026 at 15:30 UTC (60 minutes)

GitHub Issue

2

Meeting Summary:

This was the 8th P2P breakout call where the team discussed client updates and technical proposals. The meeting began with updates from various teams including Prism, Lighthouse, Teku, and Nethermind, with Ashah reporting good progress on partial columns for GLOAS and a major overhaul of Prism’s peer scoring system with improved observability and metrics. Daniel from Lighthouse shared updates on fixing field ordering issues and work on enabling patches across networks, while Csaba presented two major proposals: LODAS (a distributed reconstruction system designed to reduce CPU load by 2 orders of magnitude) and fast payload diffusion techniques for large block broadcasts. The team also discussed a new EIP proposal for segmented payload diffusion that would replace the current execution payload topic with a chunked distribution approach, with Kamil noting this would need to be discussed in tomorrow’s ACD call regarding PFI implementation. Raúl then presented his work on integrating EP2P into clients, explaining how to use a single QUIC endpoint for both ETHP2P and LIP2P while maintaining backwards compatibility, and introduced a new networking specs website to consolidate protocol documentation across multiple repositories.

Click to expand detailed summary

The team discussed updates from various clients including Prism, Lighthouse, and others. Aarsh reported good progress on Prism’s partial columns for GLOAS and a refactor of its peer scoring system with improved observability metrics. Daniel from Lighthouse mentioned fixes for field ordering issues and work on enabling patches across networks, including dynamic limits for GLOAS columns. The meeting also touched on ongoing work with Teku on vulnerabilities and segmented broadcast, as well as Nethermind’s improvements to IPv6 support.

Csaba presented Lodas, a new feature designed to implement distributed reconstruction in the network by assigning specific reconstruction duties to supernodes rather than having all supernodes perform reconstruction work on every blob. He explained that this approach would reduce reconstruction load by approximately two orders of magnitude compared to the current system, as supernodes would only handle specific rows of data rather than all blobs. Csaba also discussed his implementation of PowerShell message improvements, including changes to announcements, requests, and gossip sub-message filtering to enhance performance and efficiency in the network.

Csaba presented two main topics: changes to gos-sip sub implementation and fast payload diffusion techniques. For the first topic, he discussed improvements including reaction to IHAVs, heartbeat announcements, and metadata distribution based on node subscriptions. For the second topic, he demonstrated performance gains achieved through message segmentation, coding techniques, and phase-aware forwarding, showing how distribution times improved from 4.5 seconds to under 1 second. Raúl noted that many of these concepts are already implemented in ETHP2P broadcast and offered to provide additional context and collaboration on the existing implementation.

Kamil presented a proposal for ExecutionPayloadChunks, which involves segmenting the execution payload into chunks instead of broadcasting the full message, using 64 chunks as a starting point based on Pure DAS. The proposal requires extending the EIP to include payload chunk verification against the beacon block root, with chunks being queued until the corresponding beacon block is received. Csaba noted that having the beacon block root available would be crucial for the implementation, and both agreed that implementing this without the necessary EIP and beacon block root changes would be more complex and potentially less secure.

Raúl presented a design for integrating EF B2B into clients to enable devnets and simulations with Prism, focusing on using a single QUIC endpoint across both ETHP2P and LIP2P while maintaining backwards compatibility. The design involves splitting connections into ETHP2P and LIP2P views when both protocols are supported, using stream directionality and protocol selectors to route traffic appropriately, and managing connection lifecycle through shared transport listeners. Raúl also introduced a new initiative to create a unified repository for networking specifications to consolidate dispersed content from various sources into a single accessible location.

Raúl explained the benefits of using P2P over Libp2p, highlighting features like data grams, 0-RTT session resumption, and protocol-awareness that are not well supported in Libp2p. He expressed concerns about investing time in adapting pre-existing APIs when Ethereum-specific needs might be better served by building directly on the transport layer. Raúl emphasized the importance of focusing on reconciliation protocols rather than spending time on complex mechanisms that may not be widely adopted.

Next Steps:

  • ashah: Start an internal review of the peer scoring system overhaul PR for Prism tomorrow.
  • ashah: Conduct one more round of cortosis testing to ensure the interrupt works fine before enabling the fix on Glemshill and DevNet 8.
  • Daniel: Check the protocol-level adjustments made by Csaba to partial messages for LODAS and evaluate their necessity for the EIP.
  • Csaba: Describe the changes made to partial messages for LODAS one by one for evaluation.
  • Csaba: Extend the EIP for segmented payload broadcast today or early tomorrow to include details on handling chunks before receiving the beacon block.
  • Raúl: Give a presentation on the inventory and mechanics of ETHP2P (EIP2P) broadcast protocol.
  • Raúl: Finalize and release the networking specs aggregation site this week.
  • Raúl: Continue working on the branch for integrating ETHP2P into Prism and prepare for devnet testing.

Recording Access:

3

YouTube recording available: https://youtu.be/MChUBq0ZT94