AI-Based TCP Performance Modelling

Different mathematical models exist for modelling TCP algorithms and interrelations between TCP and network parameters. In this research, two artificial neural network models were developed to model TCP performance of both lossless and lossy traffic flows. A mathematical base line was defined for accuracy comparison in terms of regression and MSE of estimated throughput. The presence of idle time in TCP flows was investigated and accounted for in the models, in addition to the consideration of non-standard flows and statistical outliers. Neural models developed had outperformed the mathematical modelling of TCP throughput along all stages of this research. Finally, it was suggested to revise the available mathematical model to take idle time into consideration