A model and a system for data-parallel program visualization

Parallel program visualization and debugging require new techniques for gathering and displaying execution trace and profile data. Interaction with the program during execution is also required to facilitate parallel debugging. We discuss the difficulties associated with runtime user/program interaction and how the data-parallel programming paradigm facilitates much more liberal runtime interaction than typical MIMD-based models. We present a model for data-parallel program visualization that addresses both data collection/interaction and visualization issues. We follow our model presentation with the design and implementation of a subset of our visualization model. We discuss our preliminary findings and propose future research directions.