Today's Network Operation Centers (NOC) consist of teams of network professionals responsible for monitoring and taking actions for their network's health. Most of these NOC actions are relatively complex and executed manually; only the simplest tasks can be automated with rules-based software. But today's networks are getting larger and more complex. Therefore, deciding what action to take in the face of non-trivial problems has essentially become an art that depends on collective human intelligence of NOC technicians, specialized support teams organized by technology domains, and vendors' technical support. This model is getting increasingly expensive and inefficient, and the automation of all or at least some NOC tasks is now considered a desirable step towards autonomous and self-healing networks. In this work, we investigate whether such decisions can be taken by Artificial Intelligence instead of collective human intelligence, specifically by Deep-Reinforcement Learning (DRL), which has been shown in computer games to outperform humans. We build an Action Recommendation Engine (ARE) based on RL, train it with expert rules or by letting it explore outcomes by itself, and show that it can learn new and more efficient strategies that outperform expert rules designed by humans by as much as 25%.

Femto access points (FAP) are low-power cellular base stations that are designed to be autonomously deployed by customers indoors. Due to spectral scarcity, FAPs are expected to share spectrum with underlying macrocells. Closed access refers to the strategy where only Owners of the FAP are allowed access; whereas the FAP is open to everyone under Open access. Challenges such as dead zones or excessive signaling arise when implementing these two access strategies. Cognitive ac¬¬cess control is a hybrid approach that would have the FAP first senses the environment, prioritizes different classes of users, and then reserves a portion of femtocell radio resource for Owners while distributing the remaining to Visitors. Simulation results have shown that by utilizing the proposed Cognitive access control and reserve resource dynamically with the surrounding environment, both Macro-user and Owner throughputs will improve over the macrocell-only baseline, as well as both Closed and Open access strategies.

Femto access points (FAP) are low-power cellular base stations that are designed to be autonomously deployed by customers indoors. Due to spectral scarcity, FAPs are expected to share spectrum with underlying macrocells. Closed access refers to the strategy where only Owners of the FAP are allowed access; whereas the FAP is open to everyone under Open access. Challenges such as dead zones or excessive signaling arise when implementing these two access strategies. Cognitive ac¬¬cess control is a hybrid approach that would have the FAP first senses the environment, prioritizes different classes of users, and then reserves a portion of femtocell radio resource for Owners while distributing the remaining to Visitors. Simulation results have shown that by utilizing the proposed Cognitive access control and reserve resource dynamically with the surrounding environment, both Macro-user and Owner throughputs will improve over the macrocell-only baseline, as well as both Closed and Open access strategies.

Autonomous Mesh Networks potentially allows for cheaper networks, of use for impoverished areas with poor infrastructure and little interest from service providers for expansion. The subject of wireless mesh networks is interesting for several reasons. Non-reliance, or at the very least reduced reliance on existing infrastructure and service providers gives more control of a network to the users and their communities. These kinds of networks are however conceived to be quite complex to set up, manage and maintain. The goal of this paper was to create an autonomous network without any need for infrastructure, that was relatively easy to configure, use, and performs well. The implementation technique used succeeds at reaching these goals. The script and environment that was constructed makes it easy to set up and join nodes into the network, and the network can increase and decrease in size without affecting the core functionality of the network. The implementation for automatic host discovery makes it simple for anyone with a small amount of knowledge to find and communicate with other hosts, and the network has proven to be resilient to some common ways of tampering.

Made available in DSpace on 2015-04-29T15:10:48Z (GMT). No. of bitstreams: 1 ALEXANDRE PASSITO.pdf: 3822416 bytes, checksum: 4f278e2830ed590e916983c979c90872 (MD5) Previous issue date: 2012-12-04 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Autonomous networking research to applies intelligent agent and multiagent systems theory to network controlling mechanisms. Deploying such autonomous and rational entities in the network can improve its behavior in the presence of very dynamic and complex control scenarios. Unfortunately, building agent-based mechanisms for networks is not an easy task. The main difficulty is to create concise knowledge representations about network domains and reasoning mechanisms to deal with them. Furthermore, the Internet makes the design of multiagent systems for network controlling a challenging activity involving the modeling of different participants with diverse beliefs and intentions. Such type of system often poses scalability problems due to the lack of incentives for cooperation between administrative domains. Finally, as the current structure of the Internet often prevents innovation, constructed autonomous networking mechanisms are not fully deployed in large scale scenarios. The Software-Defined Networking (SDN) paradigm is in the realm of Future Internet efforts. In the SDN paradigm, packet forwarding hardware is controlled by software running as a separated control plane. Management software uses an open protocol to program the owtables in different switches and routers. This work presents a general discussion about the integration of autonomous networks and software-defined networks. Based on the knowledge offered by this discussion, it presents a framework that provides autonomy to SDN domains allowing them to act cooperatively when deployed in scenarios with distributed management. Two case studies are presented for important open issues in the Internet: (1) the problem of mitigating DDoS attacks when thousands of attackers perform malicious packet ooding and SDN domains must cooperate to cope with packet filtering at the source; (2) the problem of network traffic management when multiple domains must cooperate and modify routing primitives. / A pesquisa em redes autônomas aplica a teoria de agentes inteligentes e sistemas multiagente em mecanismos de controle de redes. Implantar esse mecanismos autônomos e racionais na rede pode melhorar seu comportamento na presença de cenários de controle muito complexos e dinâmicos. Infelizmente, a construção de mecanismos baseados em agentes para redes não é uma tarefa fácil. A principal dificuldade é criar representações concisas de conhecimento sobre os domínios de redes e mecanismos de raciocínio para lidar com elas. Além disso, a Internet faz com que o projeto de sistemas multiagente para o controle da rede seja uma atividade intrincada envolvendo a modelagem de diferentes participantes com diversas crenças e intenções. Esses tipos de sistemas geralmente apresentam problemas de escalabilidade devido à falta de incentivos para cooperação entre domínios administrativos. Finalmente, como a estrutura corrente da Internet geralmente impede inovações, mecanismos de redes autônomas construídos não são totalmente implantados em cenários de larga escala. O paradigma das redes definidas por software (SDN) está na esfera dos esforços da Internet do Futuro. No paradigma SDN, o hardware de repasse de pacotes é controlado por software sendo executado como um plano de controle separado. Softwares de gerenciamento utilizam um protocolo aberto que programa as tabelas de uxo em diferentes switches e roteadores. Este trabalho apresenta uma discussão geral sobre a integração de redes autônomas e redes definidas por software. Baseado no conhecimento oferecido por essa discussão, é apresentado um arcabouço que provê autonomia para domínios SDN, permitindo que eles atuem cooperativamente quando implantados em cenários com gerenciamento distribuído. Dois estudos de caso são apresentados para importantes questões em aberto na Internet: (1) o problema da mitigação de ataques DDoS quando milhares de atacantes realizam inundação por pacotes e os domínios SDN precisam cooperar para lidar com o filtro de pacotes na origem; (2) o problema do gerenciamento de tráfego da rede quando múltiplos domínios devem cooperar e realizar modificações nas primitivas de roteamento de redes.