A measurement-based simulation model of a web cluster representing system and application level dynamics

We present a simulation model of a Web cluster which includes important aspects of the hardware, operating and communication systems as well as the application layer for realistic performance evaluations. We formulate the model in UML taking into account both architectural and behavioral aspects and apply detailed system-level measurements at low loads for determining parameters. Measured one-way delays are used to characterize the load on the network. We apply advanced input modeling techniques including multi-modal distributions, multiple phases, and Bézier curves for unconventional shapes to represent all quantities. We also propose a new model based on differences between successive delays for modeling the autocorrelations observed during the measurements. Several internal characteristics of the cluster which significantly affect the behavior at higher loads, including buffering, resource contention, and transport layer mechanisms, are modeled in detail in our UML simulation. The model is able to predict the quantiles of the overall delay for HTTP responses of different configurations beyond the laboratory setup. Replacing all distributions with exponential ones with same means leads to significant deviations, demonstrating the need for serious input modeling.