通信网络流量建模和加权公平队列性能分析与实践  

Traffic modeling and weighted fair queueing performance analysis and practice in telecommunications

作　　者：陈刚 夏俐 姜兆宇 彭曦 徐慧颖[5] CHEN Gang;XIA Li;JIANG Zhaoyu;PENG Xi;XU Huiying(School of Management,Guangzhou University,Guangzhou 510006,China;School of Business,Sun Yat-Sen University,Guangzhou 510275,China;Huawei Technologies Co.,Ltd.,Beijing 100095,China;Huawei Technologies Co.,Ltd.,Hong Kong 999077,China;Huawei Technologies Co.,Ltd.,Dongguan 523808,China)

机构地区：[1]广州大学管理学院,广州510006 [2]中山大学管理学院,广州510275 [3]华为技术有限公司,北京100095 [4]华为技术有限公司,中国香港999077 [5]华为技术有限公司,东莞523808

出　　处：《系统工程理论与实践》2024年第4期1335-1348,共14页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(72342006,72371253,72201072);广东省区域联合基金(2022A1515110725)。

摘　　要：通信网络流量建模和排队性能分析问题对网络服务质量管理、带宽调度优化等具有重要的意义和作用.该问题也是华为公司2019年提出的通信领域十大挑战问题之一.基于和华为公司合作的研究项目,本文主要研究路由器输出端口的网络流量建模和排队性能分析问题.区别于传统的话音业务流、高速网络业务流强度变化显著,存在相关性和突发性,需要研究表达能力更强的网络流量模型.本文首先设计了基于批马尔可夫到达过程(batch Markov arrival process,BMAP)的网络流量模型拟合算法.其次,基于路由器输出端口的网络业务流服务传输机制,本文建立了加权公平队列(weighted fair queueing,WFQ)服务机制下的BMAP/PH/1排队系统模型,推导了该网络流量排队模型的系统状态平稳分布及其排队性能指标(平均队长,时延等).最后,利用华为公司测试网络流量样本数据,对本文建立的网络流量模型及其参数拟合算法进行了数值实验,对比分析了网络流量排队性能的理论和仿真结果.实验效果证明了本文研究的网络流量模型和排队性能分析方法的有效性.The internet traffic model and queueing performance evaluation are the key issues for quality of service(QoS)management and scheduling management of bandwidth.In 2019,they are also proposed by the Huawei company as one of the ten challenging problems in telecommunication area.Based on a practical project from the Huawei company,we study the traffic modeling and queueing performance evaluation in telecommunications.Unlike the classic voice flows,the high-speed traffic flows involve some statistical properties such as the correlation and burstiness.This case activates us to study the more general traffic model.In this paper,we propose a new parameter fitting approach of the batch Markov arrival process(BMAP).Based on the service mechanism of traffic flows in routers,this paper deals with a BMAP/PH/1 queueing system under weighted fair queueing(WFQ)discipline.We derive the stationary queue length distribution and performance measures(the expected queue length and delay,etc.).Finally,the performance of our proposed fitting approach is illustrated by using the teletraffic traces testing from the Huawei company.We show the effectiveness of the proposed model and the applicability of the analysis results obtained in the study via numerical and simulation experiments.

关 键 词：排队论 马尔可夫到达过程 加权公平队列 网络流量模型 

分 类 号：O226[理学—运筹学与控制论]