Neue PARvis-Komponenten zur effizienten Performance-Analyse von SVM-Fortran-Programmen

For massively parallel systems, the well-known global address space is realized by the programming model Shared Virtual Memory (SVM). This model is the base for the SVM-Fortran programming language. To achieve a better understanding of what happens in the memory during a SVM-Fortran program run, specific events are traced in a file. The analysis of the collected data is done with the visualization tool PARvis, which offers many well-designed features for this purpose. To allow the use of PARvis in the SVM-Fortran environment, this thesis first describes the development of a driver for the trace data given in a SDDF representation. The next step adds some new visualization options to PARvis to make SVM-Fortran specific information accessible to the user. Additional performance analysis is available through the source code related view, given for SVM-Fortran by the tool OPAL. Based on the common data base (trace data) conception and development of an interface between these two tools completes this thesis. (orig.)SIGLEAvailable from TIB Hannover: RA 831(3158) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekDEGerman