Reducing the grid monitoring workload using application characteristics identification

The overhead of monitoring that increases with the increasing monitoring tasks is crucial to the effective management and efficient utilization of the cluster computers and to the performance of the monitoring system. Researches are made to reduce the overhead of monitoring by identifying the main characteristics of the application. Main factors of the application are dynamically identified by performing principal component analysis (PCA) on the fly of application execution. The set of the main characteristics is identified through matching up with the comparison characteristic set that is created by 4 categories resource intensive benchmark so that the monitoring workload is reduced. A prototype monitoring system adopting the proposed strategy is implemented. Experimental results show that the monitoring workload is decreased by 20% to 60% in hybrid applications or some resource intensive applications. Large collected data can be effectively refined and processed before inputting it to a monitoring system.