自我類化在有限來源的網路上的呈現

Abstract

近來的量測研究顯示，長範圍相關 (Long-Range Dependence) 的自我類化 (Self-Similar) 程序較適合用來模擬電腦網路訊息流量，因此現今的網路研究，莫不傾向於使用自我類化程序。在自我類化的研究上，長久以來一直揣測，流通封包長度的重尾 (Heavy-Tail) 統計特性與網路自我類化度有密切的相關性。這項推測在最近已在「無限多個來源」的假設上被證實。 在本篇論文中，我們試圖研究同一個課題，但是放寬了無限多個來源的假設。亦即我們對於有限來源的重尾資料，如何在網路上造成自我類化現象以及其類化程度作實驗與觀察，並且比較分析與既有的無限來源假設之間的結果有何不同。

It has been shown recently that the modern network traffic is much more appropriately modelled by long range-dependent self-similar processes. This leads to a present research trend on network self-similarity. It has been long conjectured that heavy-tailed statistics in packet duration has a close relation with the degree of network self-similarity. Such a conjecture has recently been substantiated under the assumption that infinite network sources have been aggregated. In this thesis, we attempt to investigate the same problem by relaxing the infinite sources assumption. Specifically, our thesis experiments and observes how and how much self-similarity be contributed by finite number of heavy-tailed data sources. Analysis and comparison with that obtained under infinite source assumption are addressed.