The core of topology-aware scheduling is enabling the scheduling system to understand the physical connection hierarchy between GPUs, so that tasks with intensive communication can be consolidated onto hardware that is physically closer, reducing cross-domain traffic.