SDF's core team presents optimization work on Stellar's consensus and execution layers. Mainnet currently runs 5-second blocks at 200 TPS after recent SLP 7 upgrade. Laboratory tests achieved 2-second blocks with 3,000 TPS via protocol timing fixes, parallel transaction downloads, execution parallelization, and overlay networking redesign. Future target: 900-millisecond blocks at 10,000 TPS.
Anupani from the SDF core team walks through optimization work aimed at accelerating Stellar's consensus and execution layers. Mainnet currently produces 5-second blocks with 62 Stellar asset transfers per block, upgraded to support 200 TPS via SLP 7. The team has identified and tested improvements across three major areas: consensus timing (Protocol 28 for clock synchronization, CAP 83 for parallel transaction set download), execution (hardware upgrades to C6 instances, parallel processing, reducing thread contention yielding 6,000 TPS in 125 milliseconds), and overlay networking (separate Rust process using QUIC protocol reducing head-of-line blocking and increasing TPS by 66%). Combined optimizations achieve the interim goal of 2-second blocks with 3,000 TPS. Future work explores deterministic leader schedules, erasure coding, eager block execution, and WASME 2.0 to target 900-millisecond blocks at 10,000 TPS in a larger network.
