Models of traffic control in large Internet topologies, invited talk

Transient state queueing models help us to understand better the dynamics of Internet transmissions and the performance of traffic control algorithms. Fluid flow approximation, due its simplicity is popular and frequently used -- but even it, if applied to large topologies, is time and space consuming. The alternative approaches as Markov chains or diffusion approximation lead to much more complex computations. Fluid flow approximation is based on first-order differential equations, and its algorithm is based on iterative calculations on large mutually interdependent structures. In consequence, the bottleneck of the method lies not in numerical computations but in storing and selection of data. This is why we use an approach in which a database (SAP HANA) and its language are used to implement the method. The model logic is represented by ETL (Extract, Transform and Load) customizable and user-friendly processes. The numerical examples are based on a real topology having over 100 000 nodes. We may investigate and compare this way various TCP control algorithms as Vegas or Reno, as well as the influence of the introduction of energy aware algorithms to routers.