Research And Implementation Of Server Performance Monitoring Method Based On Agent

Server performance monitoring is very important in cloud computing,and thread deadlock detection and memory leak monitoring have always been the focus of research.However,in the current mainstream thread deadlock monitoring methods in the industry,there is a phenomenon that the update of the probe queue is not timely,and the update of the probe queue requires more message transmission.In the mainstream memory leak monitoring method,it has the limitation of static analysis.During the monitoring process,the heap dump file occupies too much system resources,and it needs repeated recursion and backtracking for basic types of data,which seriously limits the ability of these monitoring methods.Although scholars and institutions have proposed new monitoring methods to improve it,these improved methods still have problems with monitoring accuracy and performance,which deserve further optimization research.Therefore,this thesis studies and proposes a thread-centralized thread deadlock monitoring method and a memory leak monitoring method based on type point graph,and designs a performance monitoring system based on these two methods to solve the above problems.The main research work of the thesis is as follows:(1)Aiming at the problem of low performance and accuracy in Edge-chasing thread deadlock monitoring,this thesis proposes a thread-centralized deadlock monitoring method.This method uses the request resource waiting set to maintain the resource waiting relationship of each thread,and the in terms of waiting set storage,bloom filter is used to store resource waiting sets and the algorithm of bloom filter is studied to optimize the performance of thread deadlock monitoring and improve the accuracy of thread deadlock monitoring;The improved thread deadlock monitoring method is to wait for resources Release monitoring again,thereby reducing monitoring CPU usage.(2)Aiming at the problem of low performance and accuracy in memory leak monitoring in root reachability analysis,this thesis proposes a memory leak monitoring method based on type point graph.The monitoring algorithm determines the type points of memory leaks and improves the accuracy of memory leak monitoring;Finds the type point sets with memory leaks in a low-cost way,and then further specifically finds the type points of memory leaks to improve the monitoring time efficiency of the monitoring method;Storage type Points and reference edges are pooled in the global object header,thus reducing monitoring memory usage.(3)On the basis of the above research,this thesis uses Spring framework and Agent technology to design and implement a performance monitoring system based on an improved deadlock monitoring mechanism and memory leak monitoring mechanism,and finally takes functional and performance tests for the system.The test results shows that the performance monitoring system designed and implemented in this thesis meets the functional requirements and performance requirements.