高速網路中高相關資料流的輸入控制

Abstract

傳統上在分析網路效能時，為求簡化，我們多假設網路上的資料流是互相 獨立的，即符合卜松分配。但是實際上，我們發現不同的資料流來源間， 可能存在某些關聯，使其行為趨向一致或背離，這個現象對我們在高速網 路上的擁塞控制有很大的影響。在本論文中，我們提出發生這種關聯的可 能原因，包括主僕關係形及階層關係形，並藉由能更精確描述真實資料流 的多層到達程序來分析高相關資料流的特性。另外，我們使用矩陣幾何的 近似快速解法來比較Leaky Bucket對高相關性及卜松資料流的控制效果， 以做為改盡網路資料流控制之依據。 Ordinarily, when we want to simulate the analysis of network efficiency, we often assume the traffics in the network are independent to each other and correspond to Poisson distribution, in order to simplify our computation. But actually, the real traffics may have some correlations to make their behaviors trend to be the same or opposite. This phenomenon will have tremendous influence on the congestion control.In this thesis, we explain the possible reason of the occurrence of the correlation and moreover, it's pattern, such as top-down model and client-server model. Besides those two models, we use a model HAP, which can describe the real data flow more distinctly, to analyze the characteristic of the highly correlated data flow. On the other hand, we use the rapid Matrix-geometric solution to investigate the control effects the Leaky Bucket has on this correlated data flow.