Geo-Aware Leader Schedule
Solana proposes a geo-aware leader schedule to reduce propagation delays between consecutive block producers by clustering validators geographically, while quantifying trade-offs in latency gains and attack surface changes.
Solana’s current leader schedule assigns four consecutive slots to validators based on stake, which can result in geographically distant leaders causing tens of milliseconds of propagation delay during handoffs. The proposed geo-aware schedule modifies the order of the same random schedule by grouping nearby validators together, preserving each validator’s total leader windows but reducing latency between consecutive leaders. The adjustment introduces a trade-off: while latency improves, the schedule deviates from a fully random distribution, potentially increasing exposure to regional outages or censorship risks.
The analysis evaluates the latency gain and schedule deviation using synthetic networks derived from Solana’s 124 validator locations across six macroregions, redistributing stake and nodes to simulate varying concentrations. The study estimates latency between leaders based on fiber-optic signal travel speeds, modeling signals as moving at two-thirds the speed of light along paths 50% longer than great-circle distances. Positive values indicate the geo-aware schedule outperforms the random schedule in reducing handoff delays.
Results show the latency gain increases with the parameter γ and decreases with σ, with the most significant improvements observed at σ = 10% and γ = 50, yielding a 33.9 ms reduction in handoff delay compared to the random schedule. At σ = 10%, γ = 3 already achieves 65% of the maximum gain, while γ = 10 captures 91%, indicating diminishing returns beyond this point. The study highlights that beyond γ ≈ 10, additional leaders per bin provide minimal latency benefits.
The geo-aware schedule’s grouping of validators by region increases exposure to localized risks, such as regional network partitions or provider outages, as the network remains concentrated in one macroregion for longer periods. The analysis measures this ‘excess geographic persistence’ relative to a random schedule to quantify the resilience trade-off, emphasizing the need to balance latency gains with potential vulnerabilities introduced by geographic clustering.