An application can have many different data models and is completely stateless.
Layers run along-side application code. Neither stores any data, so adding more machines or losing them to failures is painless.
Each data model is provided by a different layer. Layers are also completely stateless.
Layers store their raw data side-by-side in the storage substrate.
Layers send reads & writes to the appropriate cluster machines. Load balancing and fault-tolerance is automatic.
Have a question? Ask us over at the Community Q&A →