A Self-adaptive Monitoring Architecture And Implementation For Clouds

With the rapid development of cloud computing technology,cloud computing has been widely used in all kinds of different fields.Due to the complexity of cloud's architecture and the unpredictablibilty of workloads,cloud monitoring plays a significant role in ensuring the high availability of the cloud platform.In order to solve the monitoring problems in cloud computing platform,a self-adaptive monitoring architecture is propsed for cloud computing.The proposed monitoring architecture for cloud computing is characterized by hierarchical,scalable and self-adaptive.In cloud computing,different resources are distributed in different layers,the hardware resources are distributed in IaaS,the middleware resources are distributed in PaaS,and the application resources are distributed in SaaS,just monitoring single layer of the cloud resources can not fully display the performance of the whole cloud system.The hierarchical monitoring architecture can realize monitoring underlying infrastructure,the middleware and high-level applications in the whole.Moreover,due to the heterogeneity of cloud,the workloads on cloud are dynamically changing,resources may join or leave at anytime,and the kind and amount of resources are rapidly increasing,all of these lead to the resources on cloud being dynamic,diverse and enormous scale,which brings about a great challenge for cloud monitoring.To solve the problem,we use microservices to create each independent component of the scalable monitoring system to realize the monitoring system to scale out and scale up flexibly.Howerver,the whole process of collecting,transmitting,storing and analyzing the huge amount of monitoring data can bring out large monitoring overhead.In order to reduce the monitoring overhead as much as possible,we make the monitoring architecture self-adaptive to balance the monitoring overhead and monitoring ability.It uses Principal Component Analysis(PCA)algorithm to describe the running status of system to make self-adaption by analyzing monitoring data during runtime,and use the analyzed result to adjust monitoring interval and to choose the best transmission strategy to push or pull the monitoring data.In this way,we can provide enough monitoring information to meet the need for accuracy and consistency of monitoring system.We design and implement the self-adaptive monitoring system according to the proposed cloud monitoring architecture.And we use a case of comprehensive disaster reduction system on CloudStack to evaluate our approach.According to the analysis of the monitoring results and monitoring overhead,the experimental evaluations show that the proposed monitoring approach is capable of being hierarchical,scalable and self-adaptive,and it can realize monitoring the cloud effectively and reducing the monitoring overhead efficiently.