摘要:服务质量（quality of service，简称QoS）是衡量Web服务好坏的重要标准，也是用户选择Web服务的重要依据.能够实时而准确有效地对Web服务进行监控，是Web服务质量保障的重要基础.为此，提出了一种时效感知的动态Web服务QoS监控方法.该方法在传统加权监控方法中融入了滑动窗口机制和信息增益原理，简称IgS-wBSRM（information gain and sliding window based weighted naive Bayes QoS runtime monitoring）.该方法以一定的初始训练样本进行环境因素权值初始化，利用信息熵（information entropy，简称IE）及信息增益（information gain，简称IG）对样本所处混沌状态的确定作用，依次读取样本数据流，计算样本数据单元出现前后各影响因子组合的信息增益，结合TF-IDF（term frequency-inverse document frequency）算法对早期的初始化权值进行动态更新，修正传统算法对监控分类的类间分布偏差问题和参数未更新问题.另外，考虑训练样本数据的时效性，结合滑动窗口机制来对影响因子组合权值进行同步更新，以消解长期累积的历史累赘数据对近期服务QoS的影响.在模拟数据集和开源数据集上的结果表明：利用滑动窗口机制可以有效摒弃历史数据的过期信息，结合滑动窗口机制实现的基于信息增益的动态权值算法能够更加准确地监控Web服务QoS，总体监控效果明显优先于现有方法.



Abstract:Quality of service (QoS) is an important criterion to measure the quality of Web services, and it is an important aspect for users to choose Web services. This paper proposes a dynamic weighting Web service QoS monitoring method based on information gain and sliding window mechanism. IgS-wBSRM initializes the environmental factors' weights with a certain amount of initial training samples. It also employs the theory of information entropy and gain to determine the chaotic state of the samples. IgS-wBSRM reads the sample data flow in sequence, calculates the information gain of each impact factor combination after the arrival of sample data unit. Then it updates the initialized weights with TF-IDF in a dynamic environment. In this way, IgS-wBSRM can correct the uneven classification problem between classes and the off-line constant problem in traditional monitoring approach wBSRM. Moreover, considering the timeliness of the training sample data, IgS-wBSRM combines sliding window mechanism to update the weights of each impact factor combination, so that it can eliminate the impact on the recent service running state that the accumulated historical data bring. The experiment results under a real world QoS Web service data set demonstrate that with the sliding window mechanism, IgS-wBSRM can abandon the expiration information of historical data effectively, and the dynamic weighting method combined with sliding window mechanism and information gain can monitor the QoS more accurately. The overall monitoring effect is markedly better than existing QoS monitoring approaches.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5488