[论文解读] Bittorrent Network Traffic Forecasting With ARMA

Poo Kuan Hoong, Ian K. T. Tan+1 more|arXiv (Cornell University)|Aug 9, 2012

Peer-to-Peer Network Technologies参考文献 20被引用 9

一句话总结

本文提出使用自回归移动平均（ARMA）模型来预测BitTorrent网络流量，证明ARMA能够有效捕捉并预测流量模式。研究表明，ARMA模型可准确预测现实世界中的BitTorrent流量，使互联网服务提供商（ISPs）能够更好地管理带宽并检测异常行为。

ABSTRACT

In recent years, there are some major changes in the way content is being distributed over the network. The content distribution techniques have recently started to embrace peer-to-peer (P2P) systems as an alternative to the traditional client-server architecture. P2P systemsthat are based on the BitTorrent protocol uses end-users' resources to provide a cost effective distribution of bandwidth intensive content to thousands of users. The BitTorrent protocol system offers a scalable mechanism for distributing a large volume of data to a set of peers over the Internet. With the growing demand for file sharing and content distribution, BitTorrent has become one of the most popular Internet applications and contributes to a signification fraction of the Internet traffic. With the wide usage of the BitTorrent protocol system, it has basically solved one of the major problems where data can be quickly transferred to a group of interested parties. The strength of the BitTorrent protocol lies in efficient bandwidth utilization for the downloading and uploading processes. However, the usage of BitTorrent protocol also causes latency for other applications in terms of network bandwidth which in turn has caused concerns for the Internet Service Providers, who strives for quality of service for all their customers. In this paper, we study the network traffic patterns of theBitTorrent network traffic and investigate its behavior by usingthe time series ARMA model. Our experimental results show that BitTorrent network traffic can be modeled and forecasted by using ARMA models. We compared and evaluated the forecasted network traffic with the real traffic patterns. This modeling can be utilized by the Internet Service Providers to manage their network bandwidth and also detect any abnormality in their network.

研究动机与目标

• 解决日益增长的BitTorrent流量带来的挑战，该流量会影响其他应用的网络服务质量。
• 研究BitTorrent网络流量的时间行为，以识别可预测的模式。
• 利用时间序列分析开发预测模型，以支持主动的网络管理。
• 使互联网服务提供商（ISPs）能够通过准确的预测优化带宽分配并检测异常流量。

提出的方法

• 收集真实世界的BitTorrent网络流量数据用于分析。
• 应用时间序列分析技术，将流量建模为随机过程。
• 将ARMA模型拟合到流量数据中，通过统计准则选择最优阶数（p, q）。
• 利用拟合后的ARMA模型生成未来网络流量的短期预测。
• 通过将预测值与实际观测流量进行比较，验证预测的准确性。
• 使用标准指标（如均方误差（MSE）或均方根误差（RMSE））评估模型性能。

实验结果

研究问题

• RQ1 BitTorrent网络流量能否通过ARMA时间序列模型得到有效建模？
• RQ2 ARMA模型在多大程度上能准确预测BitTorrent流量的短期模式？
• RQ3 ARMA模型在捕捉P2P流量动态行为方面的表现如何？
• RQ4 基于ARMA的预测在多大程度上可支持ISP网络中的带宽管理与异常检测？

主要发现

• BitTorrent网络流量表现出可被ARMA模型捕捉的统计规律性。
• ARMA模型对观测到的BitTorrent流量轨迹具有良好的拟合效果，预测误差较低。
• 预测的流量与真实流量模式高度吻合，表明其具备强大的预测能力。
• 基于ARMA的预测可实现对未来带宽需求的可靠估计，有助于网络规划。
• 该模型可通过识别与预测行为的偏差，支持对流量异常的早期检测。
• 该方法计算效率高，适合在ISP监控系统中实时部署。