Research Of Data Collection Technologies In Cluster Monitoring

At present, cluster system has become the development trend of high-performance computer architecture. However, in order to facilitate the management and maintence of the cluster while using cluster system in practice, it is required to realize the monitoring of the status of all the nodes and related equipments composed the cluster. So it has important practical significance for the development and application of cluster technologies to research the data collection technologies and design a cluster monitoring system.The technologies of remote cluster monitoring in storage cluster environment are researched, which takes Lustre storage cluster as background. And based on these, a cluster monitoring system prototype, which named ClusterMonitor has been implemented for Lustre storage cluster. The ClusterMonitor cluster monitoring system based on the CIM/WBEM ( Common Information Model/Web Based Enterprise Management )framework, and it is consist of MonDaemon, MonServer and MonGui three parts. The application of CIM/WBEM framework not only makes ClusterMonitor have a good scalability, but also allows ClusterMonitor no longer depends on the traditional database.The monitoring data collection is composed of monitoring data obtaining and pooling. In the monitoring data obtaining aspect, a monitoring data obtaining method, which is based on Linux kernel module, has been proposed. All the monitoring data can be obtained through a unified interface, and it is not necessary to analyze the text files. In the monitoring data pooling aspect, in order to meet the special needs of large-scale cluster monitoring, a binary tree self-restraining pooling protocol has been designed. While applying this protocol in cluster monitoring, the nodes can automatically find pooling agent and produce an optimized data pooling path.The feasibility of obtaining monitoring data in kernel state has been verified through the obtaining of memory information by experiment. And the quantitative analysis of the binary tree self-restraining pooling protocol showed while applying this data pooling protocol in large-scale cluster monitoring, the system can obtain good real-time performance.