Meeting Summary:
The meeting focused on updates and discussions regarding DevNets, repricings, and related technical aspects for the Ethereum ecosystem. Barnabas provided an update on DevNet 7, reporting stable performance with over 90% participation, though some nodes experienced disk space issues and a few remained offline. Maria presented a comprehensive repricing analysis, explaining changes to account test cases, state actor benchmarks, and the migration to DevNet 7, resulting in a new anchor of 75 million gas per second instead of 100 million. The team discussed proposed repricing parameters, with clients generally supporting the changes, though Basu raised concerns about the synthetic nature of state actor versus YoCamNet. The group also addressed the attestation deadline, deciding to maintain the current three-second deadline for now and gather more data before considering a reduction to two seconds. Additional discussions covered the implementation of DiskV5 and its compatibility with various clients, as well as potential optimizations for block access lists and snapsync functionality.
Click to expand detailed summary
Barnabas provided an update on DevNet 7, reporting it has been running for 20 days with stable performance, though some nodes experienced disk space issues unrelated to client performance. Mehdi confirmed that Teku-related bugs have been fixed and a new docker image was published. Barnabas indicated that DevNet 8 planning depends on the repricing discussion, targeting the new CL-spec release from Friday with updated repricing values.
Maria presented updates on repricing changes, explaining three main modifications including new account test cases, the switch to StateActor, and running benchmarks on Glamsterdam DevNet 7. She reported that account access operations became more expensive, requiring a shift from an anchor of 100 million gas per second to 75 million gas per second, while storage access costs will remain unchanged. The team discussed whether to freeze the proposed repricing numbers, with Maria noting that while this would unblock the fork, it would fix relative costs between operations and potentially impact future performance improvements.
The team discussed proposed changes to gas limits and execution parameters for the Glamsterdam fork. Maria presented performance data showing that with a 75 million gas per second anchor point, they could potentially target a 300 million gas limit, though she emphasized the need for a buffer due to worst-case scenario considerations. The group debated whether to start with a conservative 200 million gas limit or aim higher, with Barnabas suggesting they could make the fork decision now and adjust later if needed. The discussion ended with Justin seeking explicit approval from EL clients on the proposed changes, as a final decision had not been reached.
The team discussed and reached consensus on proposed changes to storage and account access pricing, with multiple clients including Geth, Nethermind, Besu, Erigon, and Reth supporting the updates. Ameziane raised a concern about the difference between YoCamNet and state actor testing methods, noting that state actor may be more synthetic than YoCamNet which uses mainnet snapshots. The team agreed to continue with the current numbers while investigating this assumption further, and Maria confirmed that updated execution specifications would be released once the numbers are finalized in the EIP.
The team discussed potentially reducing the attestation deadline from three seconds to two seconds, with Nico suggesting they could start with a lower value and adjust if needed. Toni raised concerns about measuring block propagation accurately, particularly with EPBS reducing block size. Mikhail emphasized the importance of testing on stable environments before mainnet deployment, while Enrico noted potential interactions with state parking heuristics that might need adjustment. Justin expressed a preference to maintain the three-second deadline and gather more data on mainnet before considering a reduction.
The team discussed block propagation timing issues, with potuz explaining that high min-bid settings by operators were causing delays, though 2-second targets might be achievable with proper testing. The group agreed to continue with 3-second timing for now and have PandaOps gather concrete performance numbers on block propagation and validation times. The team also discussed the DISC-V5 proposal, with Csaba reporting that testing showed no major issues and clients were being updated to disable fallback to DISC-V4. Finally, jochem-brouwer presented ideas about using block access lists for faster snap syncing, though Łukasz noted these optimizations were nice-to-have rather than critical for performance benchmarks.
Next Steps:
- Maria: Update the repricing numbers in the EIP and coordinate with Spencer or someone from SEAL to update the execution specs release.
- Barnabas: Schedule DevNet 8 with the updated repricing values and target it as a short-lived named TestNet.
- Pari: Work on getting the offline node (Rath and Prism) back online.
- Mehdi: Monitor the new Teku docker image to ensure the fixed bugs are working as expected.
- All EL clients: Add flags to disable zkVM4 and enable zkVM5 for the next DevNet/TestNet.
- jochem-brouwer: Investigate and provide feedback on the performance differences between StateActor and YoChemNet, especially regarding account access performance.
- jochem-brouwer: Agenda the discussion about EIP-1870 (engine API endpoints) in the ACDC meeting and check with Basel about any spec changes or concerns.
- PandaOps: Gather accurate data on block propagation and validation times post-fork to evaluate the possibility of lowering the attestation deadline to 2 seconds.
Recording Access: