利用反轉濾波器產生自我類化網路訊務之研究

Abstract

在過去的網路研究，一般都使用卜瓦松 (Poisson) 隨機程序來當作網路訊務系統的模型，這是建構在假設網路訊務系統的行為模式跟電話系統訊務的行為模式非常相似的前提。最近的網路訊務量測顯示，網路訊務的隨機程序應該具有長範圍相關 (Long-Range Dependence) 或是自我類化 (Self-Similar) 的統計特性，而與短範圍相關的卜瓦松隨機程序有顯著的不同。因此，如何合成自我類化網路訊務來進行現今網路訊務系統效能的模擬，逐漸成為重要的研究課題。 在本論文中，我們提出了一個利用反轉濾波器技術來合成自我類化網路訊務的方法。反轉濾波器是一個無限脈衝響應的濾波器，此用來取代花凱龍 [4] 先前所提出的有限脈衝響應的濾波器。基本上，這二種方法有相同的計算複雜度。經模擬比較反轉濾波器跟花凱龍的向前濾波器所合成的訊務的自我類化特性程度，出人意外地，我們所提出的反轉濾波器的方法僅能略微增加向前濾波器的方法的自我類化程度。

In early days, Poisson processes were commonly used as network tra±c models. It was done under the premise that the network tra±c behavior is similar to that in telephony system. Recent studies on network tra±c argue that network tra±c processes are much more faithfully modelled by using statistical models of long-range dependence or self-similarity, rather than the short range dependent Poisson. A nature need due to this trend is the synthesization of a self-similar network tra±c for simulation of network. In this thesis, we present a reverse-filter-based approach for synthesizing self-similar network tra±c. The reverse filter is an IIR (infinite impulse response) filter, as contrary to the FIR (finite impulse response) filter proposed by Hua [2]. Both approaches have the same computational complexity. Comparison between our reverse filter approach and Hua’s forward filter approach in degree of synthesis self-similarity is subsequently performed. Surprisingly, the improvement of our reverse filter approach over the forward filter approach is quite limited.