Building Empirical Models of Mobile Ad Hoc Networks
Abdelhamid Moursy, Ikhlas Ajbar, Dmitri Perkins and Magdy Bayoumi
International Symposium on Performance Evaluation of Computer and Telecommunication Systems (SPECTS 2007)
San Diego, California (USA), July 16-18, 2007
SPECTS_Summary
This work is concerned with studying the modeling performance of empirical-based modeling techniques when used to predict the performance of mobile ad hoc networks (MANETs). We investigate two types of empirical models (1) a first-order linear regression model and (2) a feed-forwarded back propagation neural network model. Separate models are developed for three response metrics (i.e., packet delivery ratio, the end-to-end delay, and the jitter) as a function of four factors (i.e., the average speed, the offered load, the network size, and the routing protocol). A formal statistical Design of Experiments (DOE) strategy is used to plan the experimental runs and data collection process. To investigate the performance prediction accuracy of each model type, we compare the predicted results generated by each empirical model with actual simulation results. We then provide a comparison of the advantages and tradeoffs of the two approaches.