臺灣博碩士論文加值系統

由於封包間具有相關性，封包流傾向成群的來。這種現象可能造成網路擁
塞，因而降低系統效率。為了改善系統效率，我們必須進一步探討具相關
性封包流的一些特性。在此篇論文中，我們使用多層次到達程序模組來測
量突發性。多層次到達程序模組是一種用來描述高相關性封包的模組。當
我們把封包流看成一個個突發去測量突發性時，如何決定突發是必須且重
要的。我們提出三種決定突發的方式來測量突發的平均個數、突發的平均
長度和突發性。藉著觀察這些特徵，我們也建議了一種混合的決定突發方
式。我們調整伺服器的佇列長度來測量封包的平均延遲，延遲的變異數和
流失以觀察具相關性封包的延遲和流失情形。我們發現多層次到達程序模
組有很糟的延遲和流失狀況。我們也畫了延遲和流失的曲線來決定如何取
捨。

Due to there exists correlation within packets, traffic tends
to arrive in bursts. This phenomenon may congest the network
and reduce the system performance. To improve the system
performance, we need to study the characteristics of correlated
packet traffic. In this thesis, we use HAP model, a traffic
model to capture correlation within packet traffic, as our
traffic source to measure burstiness, delay and loss. When
measuring burstiness, it is important and necessary to
determine a burst. if we regard traffic as consecutive burst
periods and silent periods. We propose three burst definitions
which can be used to determine a burst. To observe the mean
number of bursts, mean burst length and burstiness from these
definitions, we find their advantages and suggest a hybrid
method. We measure mean delay, delay variance and loss ratio by
varying server queue size to see the behaviors of delay and
loss patterns of correlated packet traffic. We find that HAP
traffic suffers serious delay and loss. We also chart delay and
loss curves to see thetradeoff between them.