“Computational Civil Engineering - "Intersections" International Journal

“Computational Civil Engineering 2005”, International Symposium 133. Implementation in C software of FFT algorithm to calculate the local kernel.4. Calculus for all receiving communications coefficients. We use the version nonblockingMPI_Wait.5. Same steps for calculus of Inverse Fourier Transform. For collecting dates weuse MPI_Gather.6. Compile the program and correct the errors to finalize the algorithm(MPI_Finalize)To prove the scalability of the algorithm, the parallel algorithm wasimplemented by a Linux cluster with Myrinet network, and we obtain thenumerical results as follows.3. NUMERICAL RESULTSNumber of samples - 2 power 16For 10 processorsTime of parallel processing = 0.190443 msEstimation in MFLOPs = 34.353376For 15 processorsTime of parallel processing = 0.18219 msEstimation in MFLOPs=35.299035For 16 processorsTime of parallel processing = 0.17611 msEstimation in MFLOPs=35.454276For 20 processorsTime of parallel processing = 0.16111 msEstimation in MFLOPs=35.7748384. CONCLUSIONSThe FFT algorithm was implemented in C program language, and theparallelization was realized by message passing with MPI help.The results obtained in a multithreaded implementation of FFT algorithmprove that the speed of computation increases by using a multiprocessor system.Even if the number of processors on such a system is, in general, small, the high-