In this paper we study resource management on self-similar traffic to guarantee QoS. At first, we investigate a local ATM network. Then a queuing process with the real ATM network traffic is established to explore the tail distribution under different conditions. We have found the following results: 1. The long-range dependent correlations are the dominate factors of the queuing performance. When the long-range dependent relation is broken, the tail length reduces very obviously, whereas the queuing performance is almost the same though the short-range dependent relation is broken. 2. The more buffer solution is not practical because the queue-length distribution of self-similar traffic is too long to be handled purely by more buffer solution. However, it can be potentially accompanied with large delay penalty. 3. The more capacity solution can reduce the queuing tail efficiently. To guarantee QoS for the long range dependent traffic, more capacity method proves to be effective. 4. Multiplexing is another solution to reduce the queuing tail length, especially at high Hurst parameter. In order to achieve good QoS on self-similar traffic, enough bandwidth and multiplexing are the suitable solutions. A simple call admission control scheme, which can achieve high utilization and sustain in the same time a relatively good QoS Guarantee, is proposed. The scheme takes the advantage of both the multiplexing and enough bandwidth allocation to guarantee QoS.
展开▼
