Basiswerkzeuge zur automatischen Auswertung von Apprentice-Leistungsdaten

The development of efficient applications in parallel computing is due to the complex parallel hardware architecture a great challenge to software designers. Conventional tools assisting optimization of parallel applications collect performance data of single program parts and visualize them graphically. However, detecting performance bottlenecks requires extensive knowledge and experience in software optimization. This thesis describes an exemplary implementation of tools supporting automatic performance analysis on a CRAY T3E within the framework of the KOJAK project (Kit for Objective Judgement and Automatic Knowledge-based detection of bottlenecks) at the Central Institute for Applied Mathematics. These tools consist of an extension of Apprentice by an export interface. Additionally, a database storing performance data and information about programs and their different versions is designed. A Java applet demonstrates the access to the database and implements the first steps of an automatic analysisAvailable from TIB Hannover: RA 831(3652) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekSIGLEDEGerman