以機器學習為基礎之可用頻寬測量方法

Abstract

一條路徑的可用頻寬即為連結中最小未使用的頻寬。可用頻寬的估計對於許多應用程式是相當有用的，例如繞徑選擇，伺服器選擇，管理控制等等。近年來，許多用來改進可用頻寬估計的工具被提出。其中兩個非常受歡迎的模型分別是探查缺口模型和探查速率模型。因為利用這兩個模型的工具已經變得成熟，我們把這兩個模型與統計方法結合起來，並提出一個工具。 我們工具的基本想法是我們能在不同的可用頻寬下收集許多不同的資料。 在收集足夠的資訊之後，我們使用統計方法分析這些數據，並且使用其結果估計可用頻寬。在我們的模擬過程中，我們使用兩種方法來收集數據，包括散佈，封包損失比率等等。然後我們使用SVM訓練這些屬性，並且估計可用頻寬。在這篇文章裡，我們詳細描述我們的工具，並且顯示一些結果來說明工具的準確度。

The available bandwidth of a path is determined by the link with the minimum unused bandwidth. The estimation of available bandwidth is useful for many applications, such as route selection, server selection, admission control, and etc. In recent years, there are many tools that have been proposed to improve the estimation of available bandwidth. The two most popular models are the probe gap model and the probe rate model. Since tools based on these two models have become mature, we propose a tool that combines statistical methods with these two models. The basic idea of our tool is that we can collect many different data under different available bandwidth. After collecting enough information, we use statistical methods to analyze these data and use results to estimate available bandwidth. In our simulation, we use two methods to collect data, including dispersions, packet loss rate, and etc. Then we use SVM to train these attributes and estimate available bandwidth. In this paper, we describe our tool in detail, and show some results to illustrate the tool’s accuracy.