Dynamic bandwidth allocation based on linear prediction for self-similar traffic

Abstract

Recently, self-similar and long range dependent characteristics have received significant attention in network traffic modelling, and it has been known that increasing the link bandwidth is more efficient than increasing the buffer size to meet the required QoS(Quality of Service) for self-similar traffic. So, many studies have been focusing on the estimation of the required bandwidth for self-similar traffic. However, a deterministic bandwidth allocation strategy seems not to be quite effective because of the Noah effect and the Joseph effect. Therefore, we propose the new method to allocate bandwidth dynamically according to the amount of traffic volume. In the proposed method, to avoid frequent estimation of the bandwidth, which is undesirable in the practical situation, we divide the M/G/∞ input process X(t) into two sub-processes, long time scale process $X_l(t)$ and short time scale process $X_s(t)$. For $X_s(t)$ which is Makovian, we allocate a deterministic bandwidth using the effective bandwidth and for $X_l(t)$ which varies (relatively) slowly and hence doesn``t need to be estimated frequently, the required bandwidth is estimated and allocated dynamically using the linear prediction. Throughout simulations we verify that our proposed method performs well to satisfy the required QoS.