Parallel computing is maturing to become a viable approach for increasing CFD mesh resolution and reducing turnaround times — many of todays commercial programs are already available for parallel architecture machines, and software tools exist to help with automated parallelization of irregular (unstructured) scientific problems. However, despite numerous impressive demonstrations of applications, it is still not exactly clear when and how these will really impact on the way we do turbomachinery CFD. The current paper does not even attempt to answer such a general question, but instead presents the authors experience and views obtained while developing parallel codes for both industry and in conjunction with their own research work.
We give a brief review of events leading up to the present situation, and explain why parallel processing is the accepted route to follow. We then describe our parallel unstructured code; discuss practical issues relating to parallel performance; present some sample test case calculations; and finally consider a possible scenario regarding the role and requirements for similar computations in the future.