By

Computer capacity planning is the process of monitoring and projecting computer workload and specifying the most cost-effective computing environment that meets the current and future demands for computer services. The growing complexity of database systems stimulates the development of software tools to probe and predict system performance and to help in system design. The main contributions of this thesis are a study of DBMS tuning factors influencing its performance under OLAP workload, the design and validation of a queueing network model to capture the main features of the DBMS behavior, and the use of a quantitative approach to project the DBMS performance with OLAP workload. We built a workload model for OLAP, which is based on the TPC-H benchmark, to investigate the impact of DBMS tuning factors (such as buffer pool, sort heap, prefetching and number of I/O servers) on OLAP performance. We found there are certain thresholds on buffer pool and sort heap size in a DBMS with OLAP workload. The relationships between these thresholds, database size and hardware configurations were further investigated. We propose a queueing network model (QNM) to represent the DBMS under study and give the results of preliminary experiments to validate this model. We have indicated, through discussion and example, how to modify the parameters of this model i to represent various common changes to the hardware (CPU and disk upgrade) and workload (number of users and database size). Amdahl’s law is used to estimate the effect of multiple disks on DBMS performance. A mathematical model is proposed and used to predict the disk demands for different disks