Deployment path for prefix-aware and load-aware LLM request scheduling. The llm-d Router considers prefix-cache affinity and server load metrics to avoid recomputation and hotspots.