Is it good to say that between nodes (physical machines) we mostly use MPI while inside a node we often use shared-address model?

Ah, I just saw the conclusion in slides 62: "Use shared address space programming within a multi-core node of a cluster, use message passing between nodes".

@icebear. Exactly. A pragmatic solution these days is to use a hybrid of programming models for big complex machines where different optimizations or strategies are best at different levels of the machine.