as @yetitwo mentioned, only if the entire dataset is completely independent would you want to have it be distinct instances; otherwise, one schema