Presented are new methods of loop execution time estimation for parallelized and distributed systems. The proposed solutions take account of data transfer time, data locality and synchronization of threads. In this way the methods have been adjusted to modern parallel and distributed systems, which permits to estimate execution times of loops compatible with FAN, PAR and PIPE transformations. The model-based estimates have been compared to real measurements of program loops parallelized in the OpenMP standard and adapted to distributed systems satisfying the MPI standard. The presented approach can be used for optimized allocation of tasks in multithreaded processors and in distributed systems.
