A Causal Approach to the Study of TCP Performance

摘要: Communication networks are complex systems whose operation relies on a large number of components that work together to provide services end users. As the quality these depends different parameters, understanding how each them impacts final performance service is challenging but important problem. However, intervening individual factors evaluate impact parameters often impractical due high cost intervention in network. It is, therefore, desirable adopt formal approach understand role and predict change any will performance.The causality pioneered by J. Pearl provides powerful framework investigate questions. Most existing theory non-parametric does not make assumption nature system under study. most implementations causal model inference algorithms examples usage rely assumptions such linearity, normality, or discrete data.In this article, we present methodology overcome challenges working with real-world data extend application area telecommunication networks, for which linearity do no hold. Specifically, study TCP, prevalent protocol reliable end-to-end transfer Internet. Analytical models TCP exist, they take into account state network only disregard at sender receiver, influences performance. To address point, as file (FTP), uses transfer. Studying well-understood allows us validate our compare its results previous studies.We first using traffic obtained via emulation, experimentally prediction an intervention. We then apply sent over Throughout studying other approaches analytical modeling simulation show can complement other.