Study of performance evaluation for data-intensive large-scale systems

The different applications make performance evaluation for data-intensive large-scale systems become a very important work. General test methods pursue the peak value as the final result without paying enough attention on resource utilization. However, the recent studies have proved that the behavior of resources can reflect the latent problems. In this paper we present a new method called POPE (Process Oriented Performance Evaluation), which analyses the systems combined workload characterization with resource utilization based on different node roles. It simplifies the complex system environment and its result contains much more information than before. This method has been applied to the real system with tens of nodes and it has found several key problems in design and equipments in time.