Performance measurement and analytic modeling techniques for client-server distributed systems.
A. Masud Khandker · Deep Blue (University of Michigan) · 1997
In this dissertation, we describe a methodology to develop analytic performance models for client-server distributed systems. The methodology decomposes distributed system models into submodels, analyzes the submodels using efficient algorithms, and combines the results from the submodels to predict the performance for the entire system. We apply our methodology to the World Wide Web (WWW) services in the Open Software Foundation's Distributed Computing Environment (OSF/DCE) and build several models using different degrees of decomposition. Each model is divided into one or more submodels. We use Layered Queueing Models (LQM) to represent submodels that contain layers of software servers and NetMod, a network modeling tool, to represent network submodels. We analyze the LQMs using the Method of Layers (MOL) and describe an integration technique to combine the performance estimates from the MOL and NetMod to predict the overall performance of the system. The close match between the model-estimated performance and measured system performance validates our models.