As a basic operation of signal processing, discrete Fourier transform (DFT) has a strict demand with executing speed. However, the current widely used fast Fourier transform (FFT) which can greatly accelerate DFT can't still adapt to the required speed when the data amount is tremendous. There is an awareness of that parallel schemes can be an efficient way to accelerate the FFT computing. In this paper, a parallel FFT course and its scheduling scheme in multi-machines environmentwere introduced. How to do further optimization on this scheme for the optimal accelerating rate was discussed, and some basic principles were pointed out.