Summary
•The von Neumann model allows us to be successful C programmers
•A parallel alternative is needed
•The PRAM is not a good alternative because it leads us to impractical algorithms like the O(loglog n) max
•The CTA does work because it charges for data motion and communication using the parameter l
•Capturing locality is a central benefit of the CTA
•CTA abstracts all contemporary MIMD machines
Finding a model is a balancing act in which the “right” things are modeled and everything else is ignored