Performance prediction methods and tools based on analytical models often fail in forecasting the performance of real systems due to inappropriateness of model assumptions, irregularities in the problem structure that cannot be described within the modeling formalism, unstructured execution behavior that leads to unforeseen system states, etc. Prediction accuracy and tractability is acceptable for systems with deterministic operational characteristics, for static, regularly structured problems, and non-changing environments.

In this work we present a method and the corresponding tools that we have developed to support a performance-oriented development process of parallel software. The N-MAP environment incorporates tools for the specification and early evaluation of skeletal program designs from a performance viewpoint, providing the possibility for the application developer to investigate performance critical design choices far ahead of coding the program. Program skeletons are incrementally refined to the full implementation under N-MAP's performance supervision, i.e. the real code instead of an (analytical) performance model is "engineered". We demonstrate the use of N-MAP for the development of a challenging application with extensive irregularities in the execution behavior, unstructured communication patterns and dynamically varying workload characteristics, thus resisting an automatic parallelization by a compiler and the respective runtime system, but also being prohibitive to classical "model based" performance prediction.