Lessons learned and perspectives on successful HPC software engineering and development

Massively parallel computing platforms have reached the teraflop threshold of computing power, with research groups around the world targeting a time frame of roughly 2010 to field the first production petaflop system. Successful development of software applications for these systems, as well as the current teraflop architectures, depends upon quality software engineering research and methodologies. Software engineering practices that lead to successful project completion and boost productivity need to be understood and addressed If not, the capabilities of current and future computer systems will easily outpace our ability to effectively use them. This paper discusses several "lessons learned" for parallel software development projects in the context of mid- to large-scale scientific computing development efforts. It also addresses issues that must be dealt with to facilitate effective software developments for scaling to even larger systems.