P2P Networking #1, May 20, 2026

Protocol Calls & happenings
1

Agenda

– Cell-level deltas ( @MarcoPolo )
– Sparse blobpool ( @healthykim, @kamilsa )
– Ethp2p updates ( @raulk )

Meeting Time: Wednesday, May 20, 2026 at 15:30 UTC (60 minutes)

GitHub Issue

2

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

3

Meeting Summary:

This meeting focused on client updates and technical discussions for blockchain development. Daniel provided updates on partial messages implementation, noting that global support is nearly complete pending review and that nodes running on Hudi showed promising results. Bosul discussed work on sparse block pools and addressed issues with Nethermind’s implementation of bitmaps, planning to update the Engine API specification. Csaba reported on Discovery V5 adoption across clients, noting that most are claiming support but some like Besu need additional work to support parallel stacks. Marco announced plans to enable partial columns on Hudi by default while keeping them disabled for mainnet, with Daniel confirming Lighthouse’s implementation is close to ready for release in the coming weeks. Raúl presented a proposal for batching attestation seals at source, explaining how this approach could reduce network traffic by allowing validators to pre-aggregate signatures, potentially addressing concerns about network bandwidth while being more user-friendly than traditional validator consolidation methods.

Click to expand detailed summary

The team discussed updates on the Lighthouse project, where Daniel reported that work on global support for partial messages is nearly complete and pending review. Daniel also shared findings from testing nodes with Hudi and partial messages, which showed promising results but limited network impact due to only five nodes supporting it. The team briefly mentioned an internal refactor in Rust.p2p switching back to a different protobuf deserialization stack, though Raúl noted this was not significant and mainly affected message payload size.

The team discussed issues with GetBlobs latency, with Marco explaining this is a critical issue that needs resolution before mainnet deployment. Raúl explained that the latency problem likely stems from how Go clients like Geth use KB stores where values are stored on disk and only loaded into memory when accessed, making the latency appear higher than expected. The team agreed that running a CPU profile during load testing would be the best way to diagnose the issue, with Marco planning to set up an environment to reproduce the problems and investigate the exact timings.

The team discussed implementing sparse block pools and memory maps, with Bosul noting that recent changes by Camille have improved clarity. They identified an issue with Nethermind’s implementation of bitmap ordering and agreed to specify the format in both the Engine API and EIP documentation. Daniel confirmed that implementing get blocks before functionality is on his to-do list and planned to start soon after completing partials for GLOAs. The team decided to postpone deployment to Glamstore.devnet 4 due to issues with Glove transactions and instead planned to first contact Lighthouse about implementing the updates on BAL.net.

Csaba discussed the status of disk v5, noting that while everyone claims to have it, there are some implementation challenges, particularly with Besu’s lack of support for parallel stacks. He mentioned ongoing work on cloning discovery networks and developing enhanced tools for observability in the mempool. Daniel reported that Lighthouse’s implementation for partial messages is nearly ready and will be released in the coming weeks, despite current latency issues.

Daniel proposed enabling partial columns on Hudi by default while keeping them disabled for mainnet, which Marco approved. The team discussed the low adoption rate on testnets and noted that hosted nodes would likely be updated quickly. Raúl presented a proposal for batching attestation seals at source, explaining how it would reduce network traffic and improve efficiency compared to validator consolidation. The discussion included questions about the batcher’s role, DOS vectors, and potential security implications, with Raúl clarifying that the proposal uses BLS’s properties to enable aggregation while maintaining security.

Next Steps:

  • Daniel: Complete the global support implementation for partial messages (finish self-review and peer review) and proceed to merge.
  • Daniel: Implement GetBlobs updates for Lighthouse to support the consolidated engine API, as it is on his to-do list and planned to start soon after finishing partial messages for GLOAs.
  • Daniel: Plan a release of partial messages (partial columns) for Lighthouse in the coming weeks (targeting sometime in June), with partial columns enabled by default on Hudi but disabled by default on mainnet.
  • Daniel: Check if any scrapers have stats on Lighthouse node distribution on Hudi to estimate adoption impact.
  • Bosul: Continue prioritizing making the sparse blob pool implementation more solid.
  • Bosul: Investigate the GetBlobs latency/performance issue, specifically by coordinating a CPU profile run during a load test on DevNet (in coordination with EthPandaOps or whoever runs the DevNets).
  • Bosul: Open a PR to clarify the endianness/bit ordering specification in the Engine API and the relevant EIP for the sparse blob pool bitmap.
  • Bosul: Contact Lighthouse about implementing GetBlobs before deploying sparse blob pool on the next DevNet (e.g., Glamstore DevNet 4 or a dedicated NFT DevNet similar to what was done for EIP-7070).
  • Marco: Set up an environment to connect to the DevNet, reproduce the GetBlobs latency issues on a machine he controls, and share findings/insights this week.
  • Marco: Investigate whether MMAP usage should be reevaluated given that it blocks Go routines in a hard-blocking manner.
  • Csaba: Follow up with Besu to help them implement parallel discovery stack support (DSP4 and DSP5 simultaneously) for the DiscV5 transition.
  • Csaba: Follow up with Nethermind to ensure DiscV5 is enabled on mainnet (not just testnets).
  • Csaba: Set up testnets running DiscV5 only to identify and resolve remaining issues.
  • Csaba: Share Lighthouse node stats on Hudi with Daniel and the team.
  • Raúl: Update the batching attestations at source EIP PR with the latest changes and share the link with the team.
  • Raúl: Further analyze the threat/attack model around the batch seal mechanism (whether it is necessary or can be simplified/removed).
  • Raúl: Speak with large operators to understand operational concerns regarding the batching attestations proposal.
  • Gottfried: Point a cryptographer from his team at the batching attestations EIP to review it from a cryptographic perspective.

Recording Access: