| Variable Bit Rate (VBR) video traffic, arising from scene changes, demonstrates long and short term correlation among consecutive frames. Proper traffic models able to capture such traffic characteristics have attracted the attention of researchers in the area of traffic engineering and their impact on network performance. Some of the models have captured scene changes which lead to bursty traffic. Others have captured correlation not only between frames but between Group of Picture (GOP). Proper traffic models can be used to synthetically generate traffic, which can be used as inputs to networks for performance studies on parameters such as cell loss and cell delay, etc.; Our first contribution is an improved Autoregressive Model which takes into account the periodicity in the GOP structure. A GOP starts with an intracoded frame (I-frame). After every 15 frames (in our case), an I-frame is generated. This is then a period of 15, and it repeats itself throughout the video sequence. The Autoregressive process has been used in existing works to model the Mpeg video traffic that attempts to capture the frame correlation as well as the gaussian shape of the bit rate variation. However, the autoregressive process alone does not capture scene changes. We propose an Autoregressive model of order P, AR(P) + IAP (Interrupted Autoregressive Process), to capture scene changes. We compare the model performance to that of the actual video trace, as well as the autoregressive process without scene changes. We have carried out a performance analysis of a Network Multiplexer with this new input stream.; Our second contribution, is in characterizing a relation between Correlation and Effective Bandwidth. Effective capacity as a function of correlation in several moments of video clips demonstrates interesting behavior. In several cases of analysis and measurements, it is observed that depending on the parameters of the source, effective capacity may be below or above the steady state value. Knowledge of the source parameters will aid network designers to better estimate bandwidth needs, particularly, in the transient time interval. Hence, attention has been given to such behavior and we have tried to explain the implications. |
