A theory and architecture for automating performance diagnosis

This prospectus describes research to simplify programing of parallel compu ters. It focuses specifically on performance diagnosis, the process of find ing and explaining sources of inefficiency in parallel programs. Considerab le research already has been done to simplify performance diagnosis, but wi th mixed success. Two elements are missing from existing research: 1. There is no general theory of how expert programers do performance diagnosis. As a result, it is difficult for researchers to compare existing work or fit their work to programers. It is difficult for programers to locate products of existing research that meet their needs. 2. There is no automated, adaptable software to help programers do performa nce diagnosis. Existing software is either automated but limited to very sp ecific circumstances, or in general, not automated for most tasks. The research described here addresses both of these issues. The research wi ll develop and validate a theory of performance diagnosis, based on general models on diagnostic problem-solving. It will design and evaluate a comput er program (called Poirot) that employs the theory to automatically, adapta bly support performance diagnosis. (C) 2001 Elsevier Science B.V. All tight s reserved.