Towards Efficient Traffic Engineering via Distributed Optimization in Large-Scale LEO Constellation

Space-air-ground integrated networks have the potential to offer extensive coverage beyond 5G and 6G services, as well as the cloud continuum. Managing network traffic in large-scale space-air-ground networks presents a significant challenge due to the dynamic nature of traffic demand and network topology. Additionally, the large-scale low earth orbit (LEO) constellation makes solving the traffic engineering optimization problem impractical. To tackle these challenges, we introduce the Cluster-based Elephant flow Splitting via Linear Programming (CESLP) algorithm, which can rapidly evaluate the maximum link utilization (MLU). CESLP classifies traffic flows into mice and elephant flows, and focuses on improving the splitting ratio for elephant flows. Using a divide and conquer strategy, CESLP divides the LEO constellation into multiple clusters, identifying the cluster with the most congested edge and minimizing its MLU by solving the Linear Programming (LP) sub-problems. Simulation is conducted on synthetic traffic matrices for Starlink constellation, and the results demonstrate the effectiveness of the CESLP algorithm in reducing MLU compared with other baseline routing algorithms.