Modelling and Analysis Techniques for Capacity Planning

摘要

The study examines techniques and tools of performance modelling and analysis for use in capacity planning. The study is based on three fundamental assumptions. First, the general plans of an organization must dominate the workload forecasting. Therefore, hierarchical models should be used to obtain a well-structured description of the workload. Secondly, the performance objectives must be stated from the users' point of view. Since users experience individual response times instead of the average system response-time, the performance objectives are assumed to be related to quantiles of the response-time distribution. Thirdly, model specification and workload identification are based on data that can be efficiently collected. When quantiles of response times are considered as primary performance indices, queueing network modelling and analysis are challenging. In the study, new techniques are developed both to approximate analytically and to simulate quantiles of response times. The accuracy of the approximation method developed is shown to be reasonable in the practical cases examined. The simulation techniques developed are based on sequential estimation. The length of a simulation run is not predefined, but controlled by the developed mechanisms. The decision to terminate the simulation is based on given accuracy requirements.