Predição de tráfego, usando redes neurais artificiais, para gerenciamento adaptativo de largura de banda em roteadores

• Ano de defesa: 2014
• Autor(a) principal: Oliveira, Tiago Prado
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Ciência da Computação; Ciências Exatas e da Terra; UFU
• Palavras-chave em Português: Aprendizagem profunda; Gerenciamento adaptativo; Largura de banda; Predição; Redes neurais artificiais; Series temporais; Adaptive management; Artificial neural networks; Bandwidth; Computer networks; Deep learning; Prediction; Time series; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/12581; https://doi.org/10.14393/ufu.di.2014.544

Since the formation of computer networks, a considerable increase in its complexity and size has been seen. Furthermore, there is also an increase in the rate of data trans- mission across communication links. This may cause problems of unavailability due to network congestion. The resource allocation and congestion control methods can be used to handle network congestion, however, they are complex issues and have been the subject of much study. Therefore, to attend these issues and improve Quality of Service (QoS), this work presents an algorithm for a network management system, which allocates adap- tively the router bandwidth, based on the network tra_c prediction. This algorithm is called Bandwidth Predictive Management with Neural Networks (GPLNEURO). The bandwidth allocation occurs in the router interfaces and focuses on fairness allocation, each interface receives only the necessary resource, accordingly to the predicted tra_c, i.e., the bandwidth adapts to receive the future tra_c in each interface. The proposed GPLaB algorithm uses the SNMP to monitor the network tra_c data, collecting the time series to predict the future tra_c. The bandwidth of the router interfaces is controlled through the Hierarchical Token Bucket (HTB) queueing discipline and depends of the predicted tra_c. The network tra_c prediction methods, used in this paper, were based on Arti- _cial Neural Networks. Traditional neural networks were compared with deep learning neural networks, which popularity has greatly increased in recent years. The popularity of the new deep learning neural networks has increased in several areas, but there is a lack of studies regarding time series prediction, such as Internet tra_c. The two main contri- butions of this work is a (i) comparative study of traditional neural networks and deep learning ones for Internet tra_c prediction and the (ii) adaptive bandwidth management algorithm to be used as an aid for computer network performance management.