Made available in DSpace on 2014-06-12T15:52:52Z (GMT). No. of bitstreams: 2 arquivo4786_1.pdf: 1721462 bytes, checksum: cd3726fecb794776ca337a60470ce4f1 (MD5) license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) Previous issue date: 204 / O comportamento do tráfego das redes atuais é caótico, possuindo propriedades estatísticas que incluem correlações temporais complexas. Neste cenário, um problema ainda não resolvido é o do congestionamento do tráfego. É um problema atual que tem sido agravado com o crescimento do uso de aplicações multimídia de tempo real com requisitos estritos de atraso. Apesar do aumento das taxas de transmissão dos canais de longa distância, o problema do congestionamento ainda persiste, sobretudo nas redes de acesso. A natureza não estacionária do tráfego das redes, as dependências temporais de curto e de longo alcance e os ruídos presentes no tráfego tornam difícil a tarefa de se fazer previsão de congestionamentos. Os benefícios de se prever possíveis congestionamentos em um tempo futuro curto (na ordem de milissegundos) são relevantes para os mecanismos de controle de congestionamento das atuais redes, pois evita o desperdício de recursos através de medidas de controle que vão desde o descarte de pacotes até o provisionamento de largura de banda adicional nos canais, passando pela limitação da taxa de entrada de pacotes na rede. Este trabalho propõe e implementa o mecanismo ProCon para identificar situações que podem causar congestionamentos em redes de computadores. O método usado para alcançar tal objetivo é a transformação de séries de tempo que representam o tráfego real nas redes para identificar a componente de tendência das mesmas. A idéia é utilizar wavelets, ou seja, ao invés de prever o comportamento do tráfego diretamente no domínio do tempo, modelamse as propriedades estatísticas da série no domínio wavelet e, em seguida, utilizam-se técnicas de estimação de tendências para indicar possíveis casos de congestionamento. O mecanismo proposto produziu o efeito desejado, pois detectou possíveis casos de congestionamento

Fractal traffic models capture uniquely statistical scaling characteristics and the burstiness of the real network traffic. The impact of such traffic models on existing networks has been studied extensively in the related literature and the relevance of such analysis to dimension the necessary resources and evaluate correctly the performance of these systems using computational simulation is well-known. On the other hand, the project of network algorithms already taking into account these models possesses several open issues due to the constant evolving network technologies. The design of high-speed networks considering fractal traffic models is the objective of this work and in order to illustrate the extension of the universe of possible applications, several different network scenarios were chosen: traffic monitoring applied to an intrusion detection system, traffic regulation to maintain the necessary quality of service in a domain according to some service level agreement, arriving traffic conformation at the entrance of an optical network domain and the planning of the necessary bandwidth for a satellite-based network carrying Internet traffic. In this work, important contributions were obtained for the previous enumerated scenarios: statistical process controls to monitor fractal sequences were developed; the leaky bucket, a traditional traffic regulator, was extended to be able to deal with fractal traffic; approaches to plan burst assembly algorithms, a class of traffic shaping mechanisms of optical burst switching networks, were introduced for the case when the incoming traffic is fractal and an estimative of the necessary bandwidth of a satellite-based demand-assignment multiple access system receiving fractal traffic at its entrance was obtained. All the presented results were theoretically and numerically studied. The results show the superior performance of the proposed solutions compared with existent techniques when the local traffic is fractal.

During the last decade, the enormous commercial explosion of the Internet has brought up the exhaustive use of Public Telephone Switched Network (PTSN) communication lines by computers (end-users) from all over the world. As a direct consequence of this explosion, a brand new research field, inside the context of the so-called Queuing Theory, has emerged. In fact, the actual dynamics of LAN/WAN data packets has revealed, statistically speaking, a very interesting hidden behavior. These traffic flows possess an intrinsical scale invariance property; this means that, whatever the observational time scale is, their statistical properties remains almost the same. This constitutes a core property of fractal processes. It is, by far, in sharp contrast with classical traffic assumptions, like those models based on Markovian rules. All measurements and results presented here were based on real LAN/WAN traffic traces gathered at ITA's gateway. Basically, this Thesis covers from how to identify the problem up to how to control it. Many statistical methods, to infer an unbiased and well-defined Hurst parameter are developed. Beside this, a mathematical formulism, to be used in our simulation studies, is presented. An open-loop Call Admission Control (CAC) scheme, based on that mathematical formulism is proposed. This CAC algorithm is confronted against the current ATM Forum's ABR service CAC close-loop algorithm. Furthermore, a pure stochastic simulation analysis of the proposed open-loop CAC algorithm, enhancing its advantages and drawbacks, is shown. From these simulation studies some new results have emerged. Finally, we can stand the following: the Internet explosion gave us the exceptional chance to have a real fractal queueing theory, such an issue has never seen before, however, under certain networking conditions, it can be reduced to the well-known classical Markovian-based queuing theory.