Research And Application On Low-overhead Cloud Application Performance Monitoring

Cloud Application is the key to cloud computing platform. Cloud application monitoring is the key to ensuring that the cloud application and the cloud computing platform is stable and efficient. The deployment and architectures of application for the cloud computing are complex and diverse. Traditional monitoring methods and monitoring component of cloud computing platform are focused on monitoring the use of resources, so it is difficult for them to meet the new monitoring requirements.In order to solve the problem that the monitoring technology is inefficiency and high resource-consuming, this paper proposes the scheme to reduce the overhead of monitoring, as follows。In order to solve the problem that the setting method of monitoring period is lack of flexibility and cause a waste of resources, we propose a dynamic monitoring period algorithm. In this algorithm, we design a modified Markov chain model combined with the second exponential smoothing method and the Markov chain model for analyzing and forecasting the status of data. According to the result of forecasting, the algorithm dynamically change the monitoring period and reduce the unnecessary waste of resources, such as CPU usage. The result of experiment shows that the algorithm can dynamically modify the monitoring period and effectively reduce the times of monitoring.In order to solve the problem that the load of sampling is too high, we propose a low-load dynamic instrumentation method. In this method, we propose the concept of isomorphic call chain and use the algorithm of K-means to analyze the call chain. The method can dynamically adjust the point of instrumentation and modify the sampling rate according to the status of system, thereby adaptively adjust the instrumentation. The result of experiment shows that the overhead is low when the system is in the steady state. In order to record application problems that may occur, the method appropriately increases the rate of sampling when the system is unusually overloaded.Finally, We present the application performance monitoring and analyzing system (JTangAPM). The system can monitor and analyze application performance and visualize the data. Then we emphatically introduce the design and implementation of the host monitoring agent and the application monitoring agent with the algorithm and method mentioned above. Practice shows that the system can maintain a smaller system resources consumption.