Conclusions

The research work we presented in this paper reflects our view that advances in performance technology will be a product of both innovative ideas and strong engineering. More importantly, as the performance complexity of parallel and distributed systems increases, it will be important to develop performance tools on a robust technology foundation, leveraging existing capabilities to realize more sophisticated functionality. We believe the tools described above demonstrate this result. Each is or will be implemented in a form that can be distributed with the TAU performance system. While this may go beyond what is necessary to ``prove'' a research result, it is in the application of a performance tool on real performance problems where its merit will be truly determined.