Skip to Main Content
Capacities of online services are mainly determined by the interactions between workload and the services of the application. As the complexity of IT infrastructure increases, it is quite difficult to match the capacities of various services without the knowledge of their behaviors. The challenge to the existing works is to keep the performance model consistent with the services under live workload, because the workload and application behaviors are varied greatly. Therefore, new methods and modeling techniques that explain large-system behaviors and help analyze their future performance are now needed to effectively handle the emerging performance issues. In this paper, we proposed an automatic approach to build and rebuild performance model according to services' history statuses. Based on these statuses, user behaviors and their corresponding internal service relations are both modeled, and the CPU time consumed by each service is also got through Kalman filter. The analyzed results of our model can explain the behaviors of both the whole system and the individual services, and give valuable information for capacity planning. At last, our work is evaluated with TPC-W bench mark, whose results can demonstrate the effectiveness of our approach.