A Performance Prediction Model for Database Environments: A Preliminary Analysis

Properly addressing the performance issues presented in database systems is and has been a significant technological challenge, this due to the uncontrolled fluctuation of user requests. Being able to predict the behaviour of such systems can greatly improve their performance. Several prediction methods, such as linear regression and autoregressive moving average, among others, have extensively been used to predict performance in shared environments where a workload is involved. However, not all them produce accurate predictions when the system is working under different workloads. In this paper, we present our preliminary results on exploring the accuracy of two different approaches (exact and approximate methods) used to predict the response time of a database system subject to different workloads in a controlled environment. Our results show that approximate methods present better prediction accuracy when compared to exact methods. Hence, we consider the main contributions of this work the following: (a) the results obtained from comparing exact and approximate methods, since they can be used as a basis for further works addressing similar problems, and (b) a preliminary prediction model also based on our findings.