BARROS, M. T. A. O.; http://lattes.cnpq.br/5530596957650388; BARROS, Michael Taynnan Albuquerque de Oliveira.
Resumo:
The increase in the number of applications and users in the Internet is the current infrastructure
problem which decreases the applications performance on these networks. Also, it
is a current research challenge for universities and telecommunications companies, which
focus their activities to the next generation networks as a solution to the problems previously
mentioned. As a possible contribution, in the logic layer of the network, this work proposes
the investigation of IP flows classifiers as a tool for Traffic Engineering in the Internet. This
process is responsible for conciliating the performance requirements of the network and the
traffic that passes by it. The study is held with the Differentiated Services architecture, in
which the traffic identification has a certain level of importance to its management as well as
the resource establishment for quality provision and connections establishment. Before this,
a study only with IP traffic classifiers based on machine learning is presented, which they
are referred in academic works as the best in performance. The calibration of the chosen techniques
is presented as well as an inter-classifiers performance comparison, which presents
evidence that all the set of classifiers is able to be implemented in the DiffServ. Therefore,
a performance evaluation of a IP network with DiffServ with the indicated classifiers shows
that this new approach increases, in a general way, the performance of applications transmiting
information by the network. This is the required conclusion and motivation to increase
the use of DiffServ in backbone networks.