For several decades now, parallel programming has been struggling with the question 
of  whether  inter-processor  communication  and  the  memory  hierarchy  should  be 
explicitly managed by programmers, or implicitly managed by hardware and system 
software. This talk provides a historic overview of how parallel programming models  
treated data locality and communication over the years, examines the state of the art in 
this area, and discusses new viewpoints on locality and communication management 
that reflect recent advances in parallel architectures and runtime systems.