Towards Sustainable Large Language Model Serving