近年来，随着信息科技产业高速发展，各种终端设备不断出现，如移动端的智能手机和物联网设备等。海量用户数据埋藏在虚拟网络中，促使人们研究和开发出各类推荐算法和产品，成为重要的一笔财富。在此现状下，网络基础设备的重要性日益增强。大量发展良好的互联网企业都建立了各自的机房。他们雇佣专业人员维护内部的数据中心，作为处理数据的重要枢纽。在保证网络基础设施提供高质量网络服务时，可用带宽是一个重要的度量。它指某段时间内某条网络路径中未使用的带宽大小。当网络中某条路径的可用带宽非常小甚至为零时，该路径会出现网络拥塞、延时增大和丢包等问题，导致用户体验下降、服务中止甚至发生事故。因此无论是研究网络路由策略、网络服务质量控制、流量工程，还是作为网络运营商、服务提供者或用户，都需要随时监测网络路径的可用带宽。本文围绕可用带宽的定义、特征及应用，针对现有可用带宽研究的局限性，以路径单向延迟和主动探测队列为基础展开研究。本文主要研究内容和贡献包括：?调研了可用带宽相关研究。本文详细给出了路径可用带宽定义，总结了相关研究优缺点，提出了改进思路；?研究路径发生拥塞后路径单向延迟的变化特征，提出了一种通用于大部分网络场景的可用带宽测量模型，即突发队列恢复模型。本文从理论和模拟实验中证明了该模型的正确性；?实现了一种可用带宽测量工具。本文首先研究了可能引起测量误差的各种机制，实现了针对性的去噪方法以消除干扰。然后本文基于突发队列恢复模型实现了多种算法，完成了该工具的代码实现。最后本文通过实验将其与多种工具进行了对比，展现了它高精度、高速度的优点；?研究了在可用带宽测量场景中，如何合理地结合机器学习算法以提升可用带宽测量效果。

It’s widely noted that a variety of terminal devices have appeared in the past few years with the rapid advancement of information technology industry, such as smartphones and Internet of things devices. Massive user data is buried in the virtual network world, which makes people research and develop all kinds of recommendation algorithms and products, which has become an important wealth. On this basis, the importance of network infras- tructure is increasing. A large number of well-developed Internet companies have estab- lished their own computer rooms. They employ professionals to maintain the internal data center as an important hub for processing data. When ensuring that the network infras- tructure provides high-quality network services, the available bandwidth is an important measure. It refers to the unused bandwidth of a network path for a certain period. When the available bandwidth of a path in the network is very small or even zero, there will be network congestion, increased delay and packet loss in this path, which will lead to the decline of user experience, service suspension and even accidents. Therefore, whether studying network routing strategy, network service quality control, traffic engineering, or as a network operator, service provider or user, it is necessary to monitor the available bandwidth of the network path at any time. This paper focuses on the definition, charac- teristics, and application of available bandwidth. We conduct the research based on path one-way delay and active measurement to solve the limitations of the existing available bandwidth research. The contents and contributions of this paper include:?this paper investigates the related research of available bandwidth. We give the definition of path available bandwidth in detail, summarize the advantages and dis- advantages of relevant research, and put forward our improvement ideas?•this paper studies the characteristics of path one-way delay after path congestion and proposes an available bandwidth measurement for most network scenarios, namely the burst queue recovery model. We prove the correctness of it in detail from theory and simulation experiments?•this paper implements an available bandwidth measurement tool. Firstly, we study various mechanisms that may cause measurement errors in detail and realize tar- geted denoising methods to eliminate their interference. Then we implement a va- riety of algorithms based on the burst queue recovery model and complete the code implementation of the tool. Finally, we compare it with a variety of tools through experiments and show its advantages of high precision and high speed?•this paper studies how to reasonably combine machine learning algorithms to im- prove the performance of the available bandwidth measurement methods.