Likely some cascading logic to suspend to nvdimms, then local storage, then spin down entirely if no requests are made for a certain period (with some variable

Likely some cascading logic to suspend to nvdimms, then local storage, then spin down entirely if no requests are made for a certain period (with some variable based on instantiation time I'm sure). Mix in machine learning pre-emptive loading for observed request patterns and it's pretty manageable. In order to be able to do that, and to save costly resources (memory and CPU) you need the workload to be interruptible vs flexible environnent which is persistant. In a sense this is closer to lambda but supporting native containers if I read into it correctly.