A Source Model for Vbr Video Traac Based on M=g=1 Input Processes

Statistical evidence suggests that the autocorrelation function of a compressed-video sequence is better captured by e ? p k than by k ? = e ? log k (long-range dependence) or e ?k (Markovian). A video model with such a correlation structure is introduced based on the so-called M=G=1 input processes. In essence, the M=G=1 process is a stationary version of the busy-servers process of a discrete-time M=G=1 queue. By varying G, many forms of time dependence can be displayed, which makes the class of M=G=1 processes a good candidate for modeling many types of correlated traac in computer networks. We derive the appropriate G that gives the desired correlation function e ? p k. Though not Markovian, the model is shown to exhibit short-range dependence. Poisson variates of the M=G=1 model are appropriately transformed to capture the frame size distribution of a video sequence. Using the performance of a `real' video stream as a reference, we study via simulations the queueing performance for two models: the M=G=1 model and the fractional ARIMA model 5] (which exhibits LRD). Our results indicate that the M=G=1 model is much more accurate in predicting the actual queueing performance. Only O(n) computations are required to generate an M=G=1 trace, compared to O(n 2) for a F-ARIMA trace (n being the trace length). Finally, we reexamine the variance-time test, which is typically used to detect LRD in empirical data. By applying it to synthetic realizations of the M=G=1 model, we show that this test can give misleading indications on the dependence structure of the examined data.