摘要
针对分布式环境下监控服务的决策判断准确度不够高的问题,提出了多层次多方面的监控数据框架,并将数据融合过程加入监控服务中。首先,基于发布/订阅中间件平台,设计了采集主机、进程和网络资源数据的监控服务;然后;应用阈值法和Three-Sigma法,把资源数据融合生成逻辑数据。真实环境中的实验结果表明,阈值法能及时准确检测用户定义事件,但参数设置经验性较强;Three-Sigma法无需事先确定参数,但抗干扰性差,导致虚警率较高。最后,利用趋势移动平均法分析某个进程内存使用量,得出内存周期性泄漏的结论。
In order to improve the accuracy of decision-making in monitor services of distributed environment,a new multi-level multi-aspect framework for monitor data was proposed,and a data fusion algorithm was introduced. Firstly,the resource data,including host,process and network, were collected on the platform of publish / subscribe middleware.Secondly,the resource data were fused into logical data through threshold method and Three-Sigma method. The experimental evaluation results show that the threshold method can find the user-defined events timely and accurately,but need strong empirical parameter settings. Three-Sigma method can do it without any setting,but has a poor anti-interference,resulting in a higher false alarm rate. At last,the trend moving average method was used to analyze the used memory in a process and detected that the process leaked memory periodically.
出处
《计算机应用》
CSCD
北大核心
2015年第A02期315-318,共4页
journal of Computer Applications
关键词
监控服务
数据融合
分布式环境
monitor service
data fusion
distributed environment
作者简介
陈笑梅（1989-），女，江苏南通人，硕士，主要研究方向：分布式系统监控；
仲哲卿（1992-），男，浙江嘉兴人，主要研究方向：分布计算、云监控和数据分发。
