Are expectations for parallelism too high? a survey of potential parallel users

Parallel computing has been portrayed as a technology that will enable users to solve very large and complex problems very quickly. But has this portrayal created unreasonably high expectations for potential parallel users? This paper reports the results of a survey designed to delineate the differences in how computer scientists and technical programmers approach parallel programming. Responses from 326 “potential parallel programmers” at Supercomputing ´93 are presented. The analysis provides information about the costs and benefits associated with parallel programming, application development activities, and the extent of tool use in those activities. It shows that users expect very dramatic gains in performance from parallel machines, while underestimating the amount of effort that will be needed. The results should be of interest to parallel tool developers and people involved in training parallel programmers