- Clients updates
- RowDAS by Csaba
- Fast payload broadcast by Csaba
- EIP-8411 Fast payload broadcast by Kamil
- [tentatively] Ethp2p progress by Raul
P2P Networking #008
Transcript
- read.ai meeting notes
Sergi added read.ai meeting notes to the meeting. Read provides AI generated meeting summaries to make meetings more effective and efficient. View our Privacy Policy at https://www.read.ai/pp Type "read stop" to disable, or "opt out" to delete meeting data.
- read.ai meeting notes
Read isn't recording yet. The Zoom host needs to grant recording permission in the meeting dialog.
- ashah
- ashah
- ashah
- Raúl Kripalani
great work Aarsh!
- ashah
Reacted to "great work Aarsh!" with ❤️
- ashah
Thanks Raul !
- read.ai meeting notes
Read wasn't granted recording permission, so no part of this meeting was recorded. Read is now leaving.
- Raúl Kripalani
this is how ethp2p broadcast works btw
- ashah
- Kamil Salakhiev
Also curious to see waggle branch once it is public🙂
- Daniel Knopik
I'd be really happy to discuss and help with a waggle impl for this once i made more progress with batched attestations :)
- Kamil Salakhiev
Reacted to "I'd be really happy to discuss and help with a waggle impl for this once i made more progress with batched attestations :)" with 🔥
- Raúl Kripalani
@Csaba did you try visualizing this in https://netviz.ethp2p.dev/ ?
- Raúl Kripalani
if not i can help with that
- Raúl Kripalani
Csaba, i've thought a lot about exactly these points when designing ethp2p broadcast, and a lot of what you've presented is already implemented. in fact, you can easily plug in a noop Encoder/Decoder to simply chunk without EC. how can we align going forward?
- Yann Vonlanthen
Would be happy to learn more about both :)
- Mikel Cortes
What is the current roadmap atm? Waggle? Ethp2p? is there any clear direction?
- Pablo Deymo | Lambda
We are also interested on ethp2p (Lambdaclass)
- Csaba
Replying to "@Csaba did you try v..." Not yet. But I should try :-)
- Raúl Kripalani
Replying to "@Csaba did you try v..." this was made with claude 4.6 😄 maybe i should clean it up
- Csaba
Replying to "Csaba, i've thought ..." Just to be clear, I also think we want EC, but I can sacrifice it if that’s a blocker for Hegota
- Raúl Kripalani
Replying to "@Csaba did you try v..." but the trace schema is stack agnostic you can create decoder that maps the trace to the visual language (and agents are great at that)
- Raúl Kripalani
i would like to think that Hegotá, being over 15mo out, can adopt ethp2p broadcast
- Raúl Kripalani
the risk is also extremely contained (and i'll explain why next)
- Raúl Kripalani
but there was no EIP submitted before the deadline (i missed it because i was looking at an old post 🙁)
- Csaba
Replying to "the risk is also ext..." I agree, fully. It is the “if others think that’s too much” part :-)
- Raúl Kripalani
that said, there's a world in which it is non-breaking for consensus (therefore technically no EIP needed)
- Raúl Kripalani
tbd honestly i'm not super plugged into Eth politics and red tape these days
- Raúl Kripalani
i have a quick presentation too on some ethp2p stuff
- Pablo Deymo | Lambda
We have a WIP implementation in Rust: https://github.com/lambdaclass/ethp2p-rs
- Kamil Salakhiev
Reacted to "We have a WIP implementation in Rust: https://github.com/lambdaclass/ethp2p-rs" with 👀
- Daniel Knopik
I have to leave on time today. Thanks for the presentations everyone!
- Csaba
This is the stream equivalent of what we do for discv4/discv5 dispatch at the datagram level … streams definitely sound more complex ;-)
- Yann Vonlanthen
Apologies, I have to head out too, thanks everyone!
- Csaba
I think we could evaluate one-by-one where breaking the existing API is worthwhile.
- Anton Nashatyrev
Can’t we just upgrade libp2p to what ethp2p provides? I.e. unidirectional streams, what else?
- Anton Nashatyrev
Replying to "Can’t we just upgrad..." I guess also datagram streams?
- jxs
Replying to "Can’t we just upg..." there's a PR for it: https://github.com/libp2p/specs/pull/680
- Anton Nashatyrev
Reacted to "there's a PR for it: https://github.com/libp2p/specs/pull/680" with 👍
- Raúl Kripalani
At a glance
Detailed notes
Client Updates
- •Prysm: partial columns PR for Gloas fully reviewed/tested, awaiting one upstream PR before merge to release branch00:06:30
- •Prysm: peer scoring overhaul dev-complete; tested on Hoodi for one week, no regressions; adds per-agent observability and pruning metrics00:07:11
- •Lighthouse: fixing incorrectly ordered fields in partial messages; Gloas column gossip limit replacing static 8MB cap with dynamic limit based on blob schedule max00:09:52
- •Nethermind: sparse blob pool in master; working on IPv6 support, near completion00:14:00
Ethp2p Progress
- •Raúl implementing shared QUIC transport for Prysm: single endpoint serves both LibP2P and ETHP2P streams; stream routing via ALPN negotiation + first-byte slash detection; backwards-compatible with LibP2P-only peers01:26:38
- •ETHP2P payload broadcast branch hooked into Glamsterdam in Prysm; ready for DevNet testing01:44:35
- •Raúl previewed unified Ethereum networking specs site aggregating consensus-specs, EIPs, and ETHP2P/Waggle specs in one taxonomy; unreleased, targeting this week01:45:19
- •ETHP2P advantages over LibP2P: unidirectional streams, datagrams, zero-RTT, session resumption, per-chunk stream interruption, slot-scoped cache lifetimes; lambdaclass has WIP Rust implementation01:51:06
Rowdas Proposal
- •RowDAS adds row-based reconstruction duties to PeerDAS: vertical (distributed CPU across supernodes) and horizontal (supernodeless fallback) reconstruction00:15:44
- •Each supernode assigned a row duty (no custody requirement); duty-sharded variant skips row topics entirely, using only column topics — simpler first-stage option00:19:38
- •KZG reconstruction per blob ~162ms single-core; RowDAS reduces per-node CPU ~2 orders of magnitude; degrades to PeerDAS levels under adversarial conditions00:22:51
- •Implemented on Csaba's Prysm branch (on top of partial messages); row-based reconstruction without supernodes confirmed working at 65+ nodes00:23:52
Eip Proposals Hegota
- •EIP-8411 (fast payload broadcast): replaces execution-payload topic with execution-payload-chunks; 64 segments, Merkle inclusion proof per chunk, payload-chunks-root added to bid; no erasure coding in v1 for simplicity01:14:44
- •EIP-8411 requires consensus change (bid field extension); without it, mixed-mode operation is possible but increases attack surface and buffering burden01:16:00
Fast Payload Broadcast
- •1MB payload over 50 Mbit uplink nodes (min-spec): current gossip-sub delivers last node in ~5s; all segmentation variants (A/B/C) converge near ~1s00:47:11
- •Variant A (normal gossip-sub + 32KB segments): 1.7s → 1.0s with batch publishing → further gains with phase-aware forwarding and disciplined pull (one requester per segment)00:50:13
- •Merkle commitment in the bid required to authenticate segments without full block; enables immediate per-segment forwarding upon receipt of signed beacon block00:52:09
- •ETHP2P broadcast already implements chunking, push→pull phase transition, and pluggable nil encoder; Raúl offers to onboard collaborators and align with Csaba's work01:03:02
Action items
- •Csaba to document partial-messages changes one-by-one for cross-team review and Waggle/ETHP2P alignmentCsaba
- •Csaba and Raúl to align benchmarks and share code between simulation frameworks (Plisma vs ETHP2P)Csaba, Raúl Kripalani
- •Kamil to extend EIP-8411 with chunk-before-bid handling rules and payload-chunks-root fieldKamil Salakhiev
Agenda
From the call issue #2214ShowHide
Key decisions
ETHP2P broadcast integration into Prysm: DevNet testing imminent on Glamsterdam payload broadcast
GlamsterdamETHP2P payload broadcast integration is ready for testing on the Glamsterdam DevNet.
EIPs discussed
AI Disclaimer: Some content or metadata on EIPsInsight may be AI-inferred or automatically compiled. If you find any discrepancy, please contact us at dev@avarch.org.