– Clients updates – Networking specs (Felix) – XMSS signatures broadcast simulations (Kamil) – 8411, waggle discussions
P2P Networking #011
Transcript
- FLCL (Nethermind)
yep please
- vlad
https://github.com/gattaca-com/seacast-rs very much work in progress
- vlad
(also spec is not finalised so everything liable to change)
- cayman
I won't be able to make it 5 hours earlier
- Matthew Keil
I can’t either for another 2 months
- Csaba
Speaking of the eth specs: it would be good if people would stop imagining eth/version for their EIPs.
- Derek Chiang | Ethlabs
Reacted to "Speaking of the eth specs: it would be good if people would stop imagining eth/version for their EIPs." with 🤣
- Csaba
Replying to "Speaking of the eth ..." We already have like 3 eth/73 EIPs
- jxs
in CL we have introduced protocol changes, i.e for gossipsub, like IDONTWANT/Partial Messages, we have done it by being retrocompatible
- Csaba
Replying to "Speaking of the eth ..." Same for writeups. Please try to avoid taking names that we would reserve for protocols we agree on. It just makes talking about these things 100x more difficult.
- FLCL (Nethermind)
Reacted to "https://github.com..." with ❤️
- cayman
I think you'll want to get more buyin on the CL side. Moving from fork-based to versioning will be a big change. Same with diff vs full-reified specs. If this is going to be the canonical spec -- even more so getting buyin before pushing thins.
- cayman
I mean, there are are a cohort of CL devs that are very active on the CL specs, we need to get those people on board especially. Otherwise its just asking for some backlash.
- jxs
Reacted to "I mean, there are ..." with 👍
- FLCL (Nethermind)
can be eth/next instead of eth/N that gets a version when work is finished
- cayman
CL specs currently has _features directory for these sorts of things
- cayman
how do you remove DoS protection from consensus specs? the gossip validation functions seem intrinsically linked to consensus logic
- cayman
The CL specs will just need to reference network version.
- FLCL (Nethermind)
same port btw?
- FLCL (Nethermind)
qmux vs yamux?
- jxs
waggle builds on top of libp2p
- jxs
the curent working versions are built on top of multistream-select
- cayman
I like the QUIC-only vision. Having a generic p2p library substrate sucks and is limiting long-term. I hinted at this, and the silver guys seemed to also. But libp2p-compatible does not necessarily mean "generic libp2p library substrate".
- cayman
I think the networking monorepo is nice :)
- vlad
Reacted to "I like the QUIC-on..." with 👍
- vlad
Replying to "I like the QUIC-on..." Silver is QUIC only
- Bosul Mun
- cayman
Only hiccup re networking-specs I see is exactly where the boundary is between CL specs and this repo. Looks like gossip validation rules are already in CL specs. Maybe also the DoS prose regarding the topics should be in the CL specs, maybe worth analysing how often these sentences get edited over time and "old versions" get updated in place.
- cayman
Reacted to "Silver is QUIC only" with 👍
- cayman
Replying to "I like the QUIC-only..." same for lodestar's new networking stack
- FLCL (Nethermind)
This is really cool! Make possible kinda in-browser light clients. Does it kinda move discovery to eth side when such web peers are involved? We can re-think les protocol considering such update. Maybe something around quic tickets? Is it on any devnet?
- Csaba
I unfortunately have to drop from the call soon. In case 8411 is discussed, FYI I’m preparing a PR to clarify the chunking descriptor and a Reed-Solomon scheme. It will be up soon.
- Mikel Cortes (ProbeLab)
is there any link to the spec @bosul ?
- FLCL (Nethermind)
Do you guys plan to test it on some devnet?
- Bosul Mun
Replying to "is there any link to..." https://github.com/ethereum/devp2p/pull/276
- Mikel Cortes (ProbeLab)
Reacted to "https://github.com..." with ❤️
- FLCL (Nethermind)
Something like ordering by peerid hash is good for reconnections also
- FLCL (Nethermind)
Wondering if dynamic regulation of such parameters might work
- andrei
What is wagle?
- jxs
via metadata Mikel
- jxs
Replying to "What is wagle?" https://github.com/libp2p/specs/pull/732
- jxs
you can always increase the publish and the gossip factor on waggle
- cayman
I know you're trying to get the shiniest numbers possible, but its probably worth running simulations with eip-7870 numbers as well -- or at least including some eip-7870-bound nodes in the simulation. Getting worse numbers can be a good thing -- it means we can justify smarter networking or updating our bandwidth requirement assumption.
- vlad
lo l- apparently they run ancient cpus and stake 32 eth ...
- cayman
If so, we should update the EIP? or delete it. But ignoring it is just bad
- Yann Vonlanthen
Reacted to "If so, we should update the EIP? or delete it. But ignoring it is just bad" with 👍
- Kamil Salakhiev
- Yann Vonlanthen
Replying to "If so, we should upd..." IMHO we would first need to show its not possible to do what we want to do with current numbers
- Kamil Salakhiev
- Yann Vonlanthen
Replying to "If so, we should upd..." But agree, then might be good to push for a change, but could be difficult to get through governance?
- vlad
what is the status of the eip currently?
- cayman
Replying to "If so, we should upd..." Yea that was my point wrt running sim with those numbers. Bad numbers are an objective-ish justification to change.
- FLCL (Nethermind)
ty cu!
- Yann Vonlanthen
Reacted to "Yea that was my point wrt running sim with those numbers. Bad numbers are an objective-ish justification to change." with 👍
At a glance
Detailed notes
Client Updates
- •Lodestar (jxs): researching EIP-8411 integration, exploring Waggle app/network interface boundary00:06:29
- •Prysm: beginning EIP-8411 prototype; reviewing Daniel's PR for 1-1 Waggle integration00:07:37
- •Nethermind: P2P improvements in .NET; interested in PQ-related protocols; will review EIP-8411 this week00:08:38
- •Geth (Felix): mempool improvements; working on post-quantum ENR and Discovery v5 transition plan; to post in Discord00:09:43
- •Lodestar (cayman): building new Ethereum-tailored networking stack in Zig; js-libp2p moving to maintenance mode00:11:05
- •Silver (vlad): libp2p-compatible but not libp2p; producing production-ready Rust zkast library00:12:00
- •Ethlambda: beacon client launched, presented in Cambridge; experimenting with PQ signature sizes vs BLS00:12:54
Organizational
- •Call time moving ~5h 30m earlier to accommodate Europe/Southeast Asia/Australia attendees; invite to be updated00:14:08
Webtx Protocol
- •Geth (Bosul): WebTX — new sub-protocol allowing browsers to submit transactions directly to EL P2P network via WebTransport/QUIC, bypassing RPC00:53:10
- •ENR extended with certificate hash field (QH); two-week rotation with advance announcement; identity proof via signed challenge on session open00:55:14
- •Browser connections limited: 10s lifetime, disconnect after first message, IP-based rate limiting on invalid transactions00:58:20
- •Protocol is write-only (transaction submit + result response); peer exchange planned; onion routing considered long-term01:01:32
- •WebTX merged in Geth; browser→node connections testable on mainnet today; spec PR: ethereum/devp2p#27601:11:12
Eip 8411 Waggle
- •EIP-8411 is CFI for Hegota; scope may be reconsidered pending upcoming ACDC discussions01:34:41
- •Daniel's PR (EIPs#12671) updates EIP-8411 to use Waggle; Lighthouse implementation PR#10604 available for CL teams to assess complexity01:34:58
- •Csaba: preparing PR to clarify EIP-8411 chunking descriptor and Reed-Solomon scheme01:35:28
Networking Specs
- •Felix demoed unified P2P networking spec repo merging devp2p (EL) and consensus-specs (CL) networking into one place00:17:36
- •Repo introduces legacy/current/experimental versioning; full reified specs per version; EIP-named draft files for in-progress proposals00:18:32
- •CL networking spec compiled to single up-to-date document (v6 at Gloas); fork-based versioning retained; gossip validation functions stay in consensus-specs00:23:55
- •Planned additions: Waggle and zkast under new 'broadcast' directory; QUIC transport spec; libp2p adapter spec00:25:55
- •jxs: Lodestar drove QUIC adoption but has no plans to deprecate TCP; cayman: QUIC-only vision appealing long-term; Silver is QUIC-only00:43:51
Xmss Signature Broadcast Simulations
- •Kamil ran gossip-sub simulations (~3,700 nodes, 32K validators) for XMSS/large-signature attestation propagation01:13:05
- •Reducing gossip-sub mesh degree from 8→4 cuts average duplicates from 4→2.5 with no latency regression01:20:22
- •Staggered push approach (hash peer IDs + message to assign pusher/waiter role) reduces duplicates to ~1.6; propagation latency unchanged or improved01:21:03
- •ProbeLab (Mikel): simulations reflect happy path only; adversarial/churn scenarios (eclipse, delayed forwarding) should be included in roadmap01:28:21
- •cayman: simulations should include EIP-7870 bandwidth-bound nodes; bad numbers justify updating bandwidth assumptions01:29:54
Action items
- •Review unified networking specs repo; provide feedback before Felix presents more broadlyCL client teams (especially active consensus-specs contributors)
- •Review Lighthouse EIP-8411 Waggle implementation (PR#10604) to assess integration complexityCL client teams
Agenda
From the call issue #2302ShowHide
Key decisions
Felix to share unified networking specs repo in Discord rather than bringing to ACD first; CL team buy-in required before any canonical transition
The transition to a unified networking specification repository (merging EL and CL networking) requires consensus from CL teams before being presented to the All Core Devs call.P2P call time moved 5h 30m earlier; Kamil to update protocol calendar invite
The meeting time was adjusted to better accommodate attendees in Europe, Southeast Asia, and Australia.
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.