Anza head of research and ETH Zurich professor Roger Wattenhofer alongside researcher Quentin Kniep have proposed a new geographic scheduling mechanism (SIMD-0675) for Solana that reduces simulated leader handover latency from 36.2 milliseconds to 17.0 milliseconds. The proposal relies on self-reported validator locations to group nearby block producers into consecutive turns, aiming to optimize handover speeds without altering overall stake allocation.
Key Takeaways
- 17.0 ms Handover Delay: Simulations reduced mean handover delay between honest validators from 36.2 ms to 17.0 ms, while median latency dropped from 23.4 ms to 4.5 ms.
- 10% Stake Floor: The draft organizes leader windows into 3-window bins bounded by a 10% stake floor neighborhood requirement.
- Adversarial Control Risk: An isolated attacker holding 5% of total stake in Sydney was shown to secure up to six consecutive windows.
- Open Draft Proposals: Introduced as pull requests on Sept. 29, both the scheduling draft and location-registration proposal remained open as of Oct. 7.
Grouping Leaders to Optimize Latency
Under current mechanics, Solana calculates a stake-weighted random leader schedule. The proposed two-pass design keeps each validator's original window count but reorders these opportunities into small geographic groups called bins. This handover efficiency relies on Alpenglow's fast leader handover protocol, where the current leader transmits block data directly to the next scheduled leader.
To test the system, researchers ran simulations using mainnet epoch 1038 stake distributions across 661 validators, adjusting locations with Globalping measurements. Across 108,000 leader windows per simulated epoch averaged over five random seeds, mapping nodes to RIPE Atlas metropolitan areas yielded a median handover drop from 23.4 milliseconds to 4.5 milliseconds. An alternative path arrangement within each bin was later added on Oct. 2.
Security Trade-Offs and Regional Vulnerabilities
While latency gains are substantial, the geographic sorting model introduces security trade-offs. In a simulated edge case, an attacker controlling 5% of total stake in Sydney—where no other validators operate in Oceania—was able to fill entire bins independently. Because adjacent bins can connect, the simulation recorded up to six consecutive adversarial windows.
Under the assumption of four slots per leader window and 200-millisecond slots, a standard three-window bin spans 2.4 seconds. The authors explicitly note that consecutive geographic control could heighten vulnerability to localized power outages, regional network disruptions, and targeted regional censorship compared to a randomized schedule.
Why It Matters
This proposal highlights the delicate trade-off between network execution speed and physical decentralization in high-throughput blockchains. While optimizing leader sequence by physical proximity delivers clear latency benefits, relying on self-reported validator locations creates vectors for localized control and regional single points of failure. As core developers evaluate SIMD-0675, the network must weigh whether millisecond-level throughput gains justify the security and operational risks of non-randomized leader ordering.
