Characterizing user workload for capacity planning
Jee Fung Pang · cIRcle (University of British Columbia) · 2010
With the widespread use of computers in today's industry, planning system configurations in computer sites plays an increasingly important role. The process of planning system configurations or determining hardware requirements for new or existing systems is commonly known as capacity planning among performance researchers and analysts. This thesis presents a refined capacity planning process for centralized computing system, with special attention to characterizing user workload for capacity planning. The objective is to make the entire process simpler for the computer user community, while relieving the capacity planner or performance analyst from having to rely on guesswork for the user workload performance factors. The process is divided into four phases; namely, data collection, data reduction, workload/user classification and, modeling and performance analysis. The second and third phases are collectively known as user workload characterization. The main objective of our workload characterization is to avoid any guess work on the performance factors that cannot be easily measured. The results of the workload characterization process are specifically meant to be used in analytic and simulation modeling. Three software tools required for the data reduction, workload/user classification and performance analysis phases have been developed and are discussed in the thesis.