Solana Proposes Leader Scheduling Changes to Reduce Block Handoff Delays
Written with artificial intelligence.
The Solana Foundation is exploring a new method to minimize block handoff delays among validators by rearranging the order of leaders. The proposed improvement, known as SIMD-0675, has shown potential to reduce delays by an average of 29.1 milliseconds in simulations, though it also increases the time that leaders from the same region consecutively produce blocks.
Proposed Changes to Leader Scheduling
The Solana Foundation is investigating adjustments to the leader scheduling process to enhance the efficiency of block generation among validators. The method involves having nearby validators take turns creating blocks, which has been shown in simulations to reduce average handoff delays by 29.1 milliseconds. However, this adjustment results in leaders from the same region being consecutively active for about 4.3 seconds longer.
Details of the SIMD-0675 Proposal
The analysis released by the Solana Foundation on October 14 evaluates the speed and security impacts of the SIMD-0675 improvement. This proposal seeks to maintain the opportunity for each validator to generate blocks while rearranging the order so that adjacent validators serve as leaders in succession. Currently, the leader schedule is determined by the stake of the validators but does not take their physical proximity into account. When leaders are far apart, data transfer can experience delays of several milliseconds.
Simulation Results
The foundation created a virtual network of 6,250 nodes based on 124 validator positions across six regions to compare results. When the 'RUN_LENGTH' is set to 7, the average reduction in handoff delay was 29.1 milliseconds, with an 86% improvement observed when the bundle length was set to 50. The simulation results are based on calculations of location and distance rather than real-world mainnet processing speeds. Additionally, the analysis highlights that with increased regional continuity, the number of consecutive leaders from the same region increases by an average of 5.4 leader intervals when compared to a random schedule.
Concerns and Further Proposals
The analysis also raised concerns about validators potentially misreporting their locations to take advantage of the system. A related proposal, SIMD-0674, suggests that validators register their physical locations with their voting accounts. The foundation has considered a stake floor of approximately 10% and a RUN_LENGTH range of 3 to 7 as potential criteria for these locality-based bundles. Adjusting the stake requirements may also impact the risk of false location reporting. The discussions around SIMD-0675 are still in a public proposal and research phase, addressing both the improvements in handoff delays and the associated security implications of regional clustering.