Design a distributed training system for deep learning models, covering data and model parallelism, scheduling, weight syncing, and failure handling.