第7卷第2期 智能系统学报 Vol.76.2 2012年4月 CAAI Transactions on Intelligent Systems Apr.2012 D0I:10.3969/i.issn.16734785.201111008 网络出版t地址：htp://www.cnki.net/kcma/detail/23.1538.TP.20120309.1559.005.html 双层隐马尔可夫链的突发流合成 谢逸，唐成华2，黄向农3 (1.中山大学信息科学与技术学院，广东广州510006；2.桂林电子科技大学，计算机科学与工程学院，广西桂林 541004;3.中山大学网络与信息技术中心，广东广州510275) 摘要：网络流模型被广泛用于构建网络与网络服务的测试环境，其准确性直接影响各种业务的性能评估结果及在 实际网络环境中的鲁棒性.随着电子商务及新型网络应用的普及，突发流现象已经成为现代互联网的主要特征之 一.针对平稳网络流而设计的传统网络流模型已经难以有效地描述现代网络中突发流的时间结构性及统计属性，从 而不能准确反映现代网络流的行为特征.为此，提出一种新的结构化双层隐马尔可夫模型用于模拟实际网络环境下 的突发流，并设计了有效的模型参数推断算法及突发流合成方法.该模型通过结构化的2层隐马尔可夫过程描述突 发流并实现仿真合成，使合成流可以重现实际突发流的时间结构性、统计特性及自相似性.实验表明，该模型可以有 效合成突发流 关键词：隐马尔可夫模型；合成；突发流；网络 中图分类号：TP30文献标志码：A文章编号：1673-4785(2012)02-010807 A doubly hidden Markov model for synthesizing bursty workloads XIE Yi',TANG Chenghua2,HUANG Xiangnong' (1.School of Information Science and Technology,Sun Yat-Sen University,Guangzhou 510006,China;2.School of Computer Sci- ence and Engineering,Guilin University of Electronic Technology,Guilin 541004,China;3.Network and Information Technology Center,Sun Yat-Sen University,Guangzhou 510275,China) Abstract:Network traffic models have been widely used to build the test environment for networks and network services.Their accuracy directly impacts the performance evaluation results of various services and their robustness in the actual network environment.With the popularity of e-commerce and new network applications,the burst traf- fic phenomenon has become one of the main features of the modern internet.Traditional traffic models designed for stationary network traffic have difficulty in effectively describing the temporal structure and statistical properties of burst traffic of modern networks,which causes them not to be able to accurately reflect the actual network environ- ment.In this paper,a new structural doubly hidden Markov model was proposed to characterize the practical burst traffic in a real network environment.Efficient algorithms for inference of model parameters and synthesis of the burst workload were also introduced.Based on the hierarchical structure,the proposed model can reproduce the similar temporal structure,statistical properties,and self-similarity of the real burst traffic.The proposed model in- cludes two hidden Markov processes.The parent Markov state process was used to describe the large-scale trends or phases of burst traffic.The child Markov process was used to describe the small-scale fluctuations that happen dur- ing a given phase of the arrival process.Experiments were implemented to validate the proposed model. Keywords:hidden Markov model;synthesize;burst workload;network 网络流模型一直是互联网研究中的一个重要 领域.一个有效的流模型是实现性能测试、容量规划 及攻击防御的基础.没有正确的网络流模型，将难以 收稿日期：2011-11-14.网络出版日期：20120309 准确预测网络性能指标的变化，从而无法使网络及 基金项目：国家自然科学基金资助项目(60970146)；教育部博士点专 项基金资助项目(20090171120001)；中央高校基本科研业 其业务实现自适应调整.因此，网络流模型对互联网 务费专项资金资助项目(11P38). 通信作者：谢逸.E-mail:xieyi5@mail.sysu.edu.cm. 及各种业务的研究与设计都具有重要的意义·