This starts to blur the line in my head between the low-level system concepts that your compiler/OS has to think about and the concepts you have to think about

This starts to blur the line in my head between the low-level system concepts that your compiler/OS has to think about and the concepts you have to think about for distributes systems. With the hardware systems and interconnects today, we've drawn a line in the sand based on what we've seen work and fail. Something like Redis is fairly well understood for many architectures. But when you have crazy NUMA architectures and different memory models like Intel Optane, what changes? Where do you draw the "new" line for when to use Redis vs keep everything "local" to one box? Will we see a local Kubernetes extension (or new software) that makes building for "massive boxes" easier? Kind of a cool idea.