Internet Performance: TCP in Stochastic Network Environment

The analysis of Internet traffic behavior under the pressure of long range dependence congestion factors has a great importance for performance management, anomaly detection and QoS control. The choice of methods using for analysis depends on the goal of description, accuracy requirements, measurement tools and assumptions concerning stochastic feature of network processes. Analysis without robust experimental data will fail to unrealistic result that does not reflect specific network characteristics. Proposed model describes macroscopic long-term behavior and fractal nature of congestion factors. The offered approach is proved by means of comparison of experimental and modeling results for a wide range network condition. Obtained estimations follows from the microscopic or transaction analysis of Reno congestion avoidance algorithm and user-oriented performance metrics.