自适应大规模服务器集群监控系统的构建

Design and Implementation of an Adaptive Monitoring System for Large-Scale Server Clusters

针对大规模服务器集群监控系统域大小固定而不能动态适应集群规模变化的缺点,提出了自适应动态域算法.通过计算监控引入的负载,根据负载容忍度要求,动态改变域的个数和大小,以适应集群规模变化.同时,提出了浮动域监控节点的方法,以应对域监控节点负载过重的情况,即在域监控节点负载较重时,自适应地选择新的域监控节点,并完成节点角色的自动更新.采用自适应的线程池和数据库连接池可降低监控任务的开销,加快执行任务的响应时间.监控CPU利用率的测试结果表明,系统执行单任务的响应时间为128 ms,20个任务的响应时间为242 ms;数据库连接池的性能测试结果表明,在300 s内,采用了自适应池、不采用自适应池方案所耗处理器资源分别为1.80%和7.53%,处理任务数分别为12 364和2 769.

An adaptive dynamic domain algorithm has been proposed to solve the issue that the fixed domain can not adapt to the variation of the cluster scale. By computing the intrusive load obtained by the monitoring system, the algorithm can control the numbers and the sizes of monitoring domains in an acceptable level according to the intrusion tolerance degree. The fixed domain monitoring node is not efficient as its load is high. A floating domain monitoring node method has been proposed to re-select the domain monitoring node and automatically update the roles of the nodes when the load of the node is high. To reduce the overhead incurred by monitoring system and improve the response time of executing monitoring tasks, an adaptive thread pool and a database connection pool have been designed. The measurements on monitoring CPU usage show that the response time of monitoring one node is 128ms, and that of monitoring 20 nodes is 242ms. A comparative performance measurement has been performed＇ between the solutions with an adaptive database connection pool and that without a pool. The results show that the average intrusion on CPU usage in 300 seconds is 1.80% with a pool and 7.53% without a pool, and the numbers of completed tasks are 12 364 and 2 769, respectively.