Proposta de um modelo para análise de desempenho do escalonador WFQ alimentado com tráfego LRD

• Ano de defesa: 2009
• Autor(a) principal: Nabas, Kleber Kendy Horikawa
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial
• Palavras-chave em Português: Teoria das filas; TCP/IP (Protocolo de rede de computação); Sistemas de telecomunicação; Telecomunicações - Tráfego; Algorítmos; Engenharia elétrica; Queuing Theory; TCP/IP (Computer network protocol); Telecommunication systems; Telecommunication - Traffic; Algorithms; Electric engineering
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/123

Internal routers of telecommunication networks employ a variety of queue management and packet discarding techniques. The development of models for analyzing performance of different protocols and devices is made difficult by the self-similar nature of network traffic. Traffic self-similarity stems from the action of predominant protocols and end-to-end congestion control systems, which are key factors in determining the behavior of data flow between layers of the TCP/IP protocol stack. This thesis proposes analytical models that allow for the evaluation of a (DiffServ, IntServ, IP, Multi Protocol Label Switching etc.) network node's performance at a low computational cost. Only TCP traffic is considered, since it accounts for most of current Internet traffic and therefore exhibits self-similar characteristics. Models were developed for the WFQ scheduler with Poisson and self-similar traffic inputs. The former is based on a bi-dimensional Markov chain representing the behavior of the WFQ scheduler with two input flows with different priorities. The latter consists in two separate queues whose service rates are tuned so as to simulate the behavior of a WFQ scheduler. M[X]/M/1/B queues are used to account for the self-similar nature of traffic. As a result, it is shown that the second model accurately estimates performance metrics for certain ranges of parameter values, as confirmed by NS-2 simulations.