Programming High Performance Parallel Machines

Many researchers are working on the problem of providing effective programming paradigms for parallel machines. Much of this research is motivated by the difficulties involved in automatically parallelizing programs that were initially written with a sequential machine model in mind. Another major motivation for the research is to provide models that are more flexible so that both the programmer and the compiler can best utilize the capabilities of a target machine. Such models may exhibit explicit and/or implicit parallelism, and offer a shared or distributed model of memory (or may offer no memory model at all). No matter what features the models exhibit, the key is that all such models take into account the potential parallel nature of a computation, so as not to constrain the execution of a program written within the model to necessarily be sequential. Unfortunately, it is extremely difficult to convince users to change their programming model without convincing them that the benefits of learning a new paradigm, such as potential performance gains and greater productivity, greatly outweigh the difficulties involved in changing the way they look at solving their problems.