PQ Interop #045
Transcript
Ask about this call
Answered from this call's summary and transcript only.
AI-generated — verify against the recording.
Call summary
Targets
- •Zeam: begin multi-client DevNets by end of this week — 00:19:34
- •Goal: run a month-long stable DevNet with 100+ (ideally 1000+) validators before end of year — 00:35:14
Decisions
- •Zeam (Anshal) will PR to revert Ethlambda's earlier vote-packing spec change, as it is now redundant following the latest-justified store PR merge — 00:20:43
- •Increase DevNet 5 slot time in the lean spec (currently 4s) to resolve proof aggregation bottleneck on commodity hardware; exact value TBD — 00:48:27
Highlights
- Client Updates:
- ·Zeam: 32-node, 4-subnet all-Zeam Shadow network running to finality; QUIC P2P multi-threaded, stabilization expected in 1-2 days — 00:16:31
- ·Ream: interoping successfully with Ethlambda on one subnet; plan to add Zeam, Lantern, and Grandine for 24-validator DevNet 5 runs — 00:23:30
- ·Lantern: avg Type 2 proof generation ~2.8s on Xeon/8-core; finality gaps when block building exceeds 10s; adding subnets and clients next — 00:24:09
- ·Ethlambda: updated attestations to use current head's latest justified as source; added flag to disable Syngate mechanism that caused non-finality; working on pre-build optimization — 00:26:12
- ·Gean: stable multi-client runs with no SSZ regressions; Shadow Network Simulator integration in progress — 00:27:58
- Testing Progress:
- ·Ethlambda: finality lags observed (~30 slots at a time) with 16 nodes on a 32-thread machine; CPU is the primary bottleneck — 00:40:28
- ·Ethlambda confirmed Shadow-compatible; Zeam will begin interop Shadow sims with Ethlambda before next call — 00:36:33
- ·Lantern confirmed its new libp2p library is Shadow-compatible; will submit a PR for Shadow simulator support — 00:40:30
- ·Root cause of finality lag identified as recursive proof aggregation during block proposal, not XMSS row signatures — 00:42:04
- Spec And Cryptography:
- ·Lean spec cleanup complete; no major changes expected; internal EF spec review meeting planned for next week — 00:30:04
- ·Zeam to PR revert of Ethlambda's earlier vote-packing optimization (now redundant after latest-justified store PR merge) — 00:30:54
- ·EF team exploring binary fields as proving system to potentially replace Poseidon in the future — 00:31:09
- ·Proposed path: increase slot time in spec (from 4s to 6-8s) to unblock scaling on commodity hardware; revisit in a few months as performance improves — 00:48:27
Action Items
- •All client teams — Post a DevNet 5 issues list to the DevNet 5 channel: include interop blockers, finality problems, validator counts, and proof aggregation status — 00:49:12
- •Lantern, Ethlambda, other client teams — Client teams wishing to join Shadow sims should implement sleep/simulated-time config at heavy computation points; contact Zeam (Parthasarathy/Kai) for config details — 00:39:00
Meeting chat22
- Mega | Lambda
gm
- Ruslan Tushov
Hello. No updates from Qlean this week.
- Gajinder Singh
yea similar to our observation on normal machines
- Mega | Lambda
Same
- alan xu
Will this affect the client teams of lean? [此版本无法显示完整消息]
- Gajinder Singh
I just want to mention that store's latest justified is still releavnt - its jused to choose the subtree for finding head/safe target whild doing lmd ghost
- Mega | Lambda
Yes. I expect we can revert the change once we swap 3sf-mini for other finality gadget
- Gajinder Singh
Replying to "Yes. I expect we can..." we will do a PR for revert of the previus PR you/ethlamda did for the packing votes before head's latest justified
- Gajinder Singh
(simplification even though the other PR is still strickly speaking "correct")
- Mega | Lambda
Replying to "Yes. I expect we can..." Sounds good to me
- Kamil Salakhiev
ethlambda is also shadow compatible
- Pablo Deymo | Lambda
Reacted to "ethlambda is also shadow compatible" with ❤️
- Mega | Lambda
Do we need to add a sleep instead of true aggregation for these simulations too? I recall Zeam having something like that
- Kamil Salakhiev
also btw I developed distributed shadow which allows super large scale shadow simulations on multiple colocated instances (if/when we hit hardware limits for running simulations on a single machine) https://github.com/qdrvm/shadow-distributed
- Mega | Lambda
Reacted to "also btw I developed..." with 🔥
- Mihir Faujdar
Confirming the new Libp2p library being used by Lantern is also compatible with the shadow simulator tool. We will be able to submit a PR for this.
- Daniel Phạm
Reacted to "also btw I developed..." with 🔥
- Parthasarathy Ramanujam
I think its related to the slow block proof production.
- Kamil Salakhiev
Reacted to "Confirming the new L..." with 🔥
- unnawut
Reacted to "I think its related to the slow block proof production." with 👍
- Mihir Faujdar
Reacted to "I think its related ..." with 👍
- T. Wambsgans
Ok i see, thanks for the details
Key decisions
Zeam (Anshal) will PR to revert Ethlambda's earlier vote-packing spec change, as it is now redundant following the latest-justified store PR merge
The decision was made to revert the vote-packing spec change due to it being redundant after the latest-justified store PR merge.Increase DevNet 5 slot time in the lean spec (currently 4s) to resolve proof aggregation bottleneck on commodity hardware; exact value TBD
The decision was made to increase the slot time in DevNet 5 to resolve the proof aggregation bottleneck on commodity hardware.