Amostragem aleatória estratificada adaptativa para identificação de fluxos 'elefantes' em redes convergentes

摘要

Adaptive stratified random packet sampling technique to identify large flows (“Elephant” flows) in the context of the convergent communication networks based on the IP model was implemented, evaluated and the obtained results compared with the results collected from traditional per-flow measurement system. The correlations and divergences diagnosis of the inferred information about precision, reliability and occurrence of false positive and false negative, also, was made. It was shown that the adaptive stratified random sampling requests the use of mechanisms specifically developed and it should be used with base in a previous knowledge of the usual network behavior. It was verified that, using the adaptive stratified random sampling technique, the percentile error for "elephant" flows was less than 3% in the estimation of packages and volume of bytes account; that the time model AR(1) for five past values makes the sampling technique truly adaptive and, for bursty traffic conditions, the time model AR(1) for three past values presents a larger convergence than the model AIR (1) for five or seven past values. This work also shows a bibliography review of the main aspects related to network management, converging to the state of art related to the application of the sampling packets technique. Additionally, the used sampling technique is presented and results achieved are discussed.