大规模网络服务系统行为异常的敏捷感知方法

发布时间：2018-06-20 22:58

本文选题：网络服务系统 + 行为异常　；参考：《计算机学报》2017年02期

【摘要】：大规模网络服务系统环境中,短时的大规模用户合法行为聚集会造成系统行为异常,使得系统可用性受到极大的损害.现有的系统异常检测方法大多适用于用户非法行为造成的系统异常.文中针对用户合法行为短时聚集引起的系统异常问题,提出一种大规模网络服务系统行为异常敏捷感知的方法.该方法包括系统异常敏捷感知模型和重复行为检测的Petri模型.基于"放大因子"的系统异常敏捷感知模型给出了系统异常的可预知性和异常系统行为的可定位性.即如果系统会在未来的某一时刻t2发生异常,那么t1时刻预期系统负载值要大于系统所能承受的最大负载值(t2t1).而且,该模型通过行为阻滞度可确定引起系统异常的系统行为.在敏捷感知模型的基础上,针对应用系统的异常检测,提出了基于带优先关系的颜色双变迁Petri网的重复行为检测模型及其系统异常敏捷感知算法.根据单位时间内用户行为数较小变化、缓慢增长和激增三种情况进行模拟实验,实验结果表明该方法可以有效地在系统异常发生之前提前感知,并能定位引起系统异常的系统行为.
[Abstract]:In a large scale network service system environment, the aggregation of the legal behavior of a large scale user in a short period of time will lead to the abnormal system behavior and greatly impair the availability of the system. Most of the existing system anomaly detection methods are applicable to the system anomalies caused by the illegal behavior of users. In this paper, aiming at the problem of system anomaly caused by the short time aggregation of user's legal behavior, a method of behavioral agility awareness of large-scale network service system is proposed. The method includes system anomaly Agile aware model and Petri model for repeated behavior detection. Based on the "magnification factor", the system anomaly agility awareness model is proposed, which gives the predictability of the system exception and the localizability of the abnormal system behavior. That is, if the system will have an anomaly at a certain time in the future, then the expected load value of the system at T1 time will be greater than the maximum load value that the system can bear. Furthermore, the model can determine the system behavior that causes the system to be abnormal by the degree of behavior retardation. On the basis of Agile perception model, an anomaly detection model based on color bitransition Petri net with precedence relationship and an anomaly Agile sensing algorithm for application system are proposed. According to the small change of user behavior, slow growth and rapid increase in the number of user behaviors per unit time, the experimental results show that the proposed method can effectively perceive the system before the anomaly occurs. And can locate the system behavior that causes the system anomaly.
【作者单位】：东华大学计算机科学与技术学院;安徽师范大学数学计算机科学学院;同济大学嵌入式系统与服务计算教育部重点实验室;
【基金】：国家自然科学基金(61472004) 上海市“科技创新行动计划”高新技术领域项目(16511100903) 同济大学嵌入式系统与服务计算教育部重点实验室(2015)资助~~
【分类号】：TP301.1

【参考文献】