Analytic modeling and measurement methodology for multi-tier client-server systems.
Yi-Chun Chu, Toby J. Teorey, Charles J. Antonelli · Deep Blue (University of Michigan) · 1998
Capacity planning in distributed environments encounters several traditional difficulties and a new challenge. These difficulties stem from the complexity in hardware configurations, the complexity in software architectures, and data deficiencies. The challenge arises because of hardware price drops and technology advances which often make simply purchasing new equipment more cost-effective than conducting capacity planning studies before purchasing. A new objective has emerged to meet these difficulties and challenges which requires a broader focus with less precision to assure the cost-effectiveness of capacity planning. This dissertation approaches the new objective with a closely coordinated approach to analytic modeling and measurement for complex multi-tier client-server configurations being implemented today. This approach relies on easy-to-use modeling tools built from proven analytic techniques for quick construction and evaluation of performance models, and lets capacity planners focus on developing appropriate measurement methodologies to obtain application oriented performance data. The motivation behind this approach is that application oriented performance measurements are currently the most challenging part in analytic modeling of complex client-server systems. The virtue of this approach is that analytic performance models can be constructed and evaluated quickly with the results serving as feedback for improving the measurement methodology. These two procedures iterate one after the other until either satisfactory results are reached or more effort is no longer considered cost-effective. Due to the broad ranges of client-server configurations implemented today, we are not able, and it is impossible, to address all these configurations. The feasibility and generality of this approach, however, have been demonstrated in this dissertation through several capacity planning studies of client-server systems including analytic modeling and measurement of communication overhead at end hosts, general two-tier database applications based on Oracle7 servers, and a popular remote computing architecture for Windows applications based on Citrix WinFrame servers.