To optimize online smoothing of real-time VBR video streaming, this paper proposes a SPSF(Shortest Path Smoothing based on Funnel) algorithm. In the SPSF, the real-time VBR video is piecewise processed by sliding window. Each video frame is read and cached into the window and the shortest path for the data transmission in the window is calculated based on funnel principle. When the window is filled with data, data start transmission according to the obtained shortest path. Window simultaneously slides according to the characteristics of the path to start smoothing and transmission of the next piece, and so complete the smooth transmission of the entire video. The experimental results show that compared with conventional online smoothing algorithm, SPSF has better bit-rate peak, valley, and variance. Compared with traditional shortest path algorithm, SPSF takes less time in computation of shortest path so as to be able to meet the real-time requirement.