Thesis (Ph.D.)--University of Wollongong, 2005. / Typescript. Includes bibliographical references: leaf 88-95.

CAPES / Neste trabalho é apresentado o desenvolvimento e análise de arquiteturas de dissipadores de calor por redes de microcanais. As configurações em forma de redes são caracterizadas geometricamente por múltiplas ramificações no escoamento e variação nas escalas de comprimento e diâmetro hidráulico através de cada nível de ramificação. O momento tecnológico atual tem permitido a fabricação e a experimentação de redes de microcanais, todavia, verificou-se que as arquiteturas investigadas experimentalmente nos últimos anos têm parâmetros geométricos constantes através de seus níveis de ramificações, o que se denomina fractal. Neste trabalho utiliza-se a teoria constructal para projetar arquiteturas de redes com geometria variável e até três níveis de ramificação. Algumas hipóteses comumente empregadas no desenvolvimento de geometrias constructais em macroescala, como por exemplo, escoamento completamente desenvolvido e resistência térmica parede- fluido desprezível, são reconsideradas por se tratar de uma aplicação de microescala. Além disso, a geração de arquitetura é feita para um microchip de tamanho e razão de aspecto definidos. Como resultado, foi verificado que as redes constructais permitem uma redução significativa na queda de pressão em relação à redes fractais com mesmos níveis de ramificação. Foi demonstrado que a rede bifurcada com razão de diâmetros segundo a lei de Hess-Murray não é apropriada para dissipação de calor em dispositivos miniaturizados. Curvas de resistência térmica versus potência de bombeamento são mostradas para evidenciar a notória superioridade das redes constructais em relação às fractais. / The present work introduces microchannel nets architectures development and analysis for heat dissipation purposes. The net configurations are geometrically characterized by multiple flow ramifications and changes in length and hydraulic diameter scales through each ramification level. The current technological state has allowed manufacturing and experimentation of microchannel nets, however, it was found that architectures investigated experimentally in the past years have constant geometric parameters through their ramification levels, which is denominated fractal. In this study constructal theory is used to design net architectures with variable geometric parameters and up to three ramification levels. Some hypothesis commonly employed in macro scale analysis, for instance, fully developed flow and negligible wall-fluid thermal resistance, are reconsidered in order to comply with micro scale applications. Moreover, the architectures design is elaborated for a chip with fixed size and shape. As a result, It was verified that constructal nets allow a significant pressure drop decrease with respect to fractal nets with same ramification levels. It was demonstrated that the bifurcated net with diameter ratio according to Hess-Murray law is not appropriated for heat dissipation in miniaturized devices. Thermal resistance versus pumping power curves are shown in order to evidence the notorious superiority of constructal nets compared to fractal.

Digital audio network technologies are becoming more prevalent in audio related environments. Yamaha Corporation has created a digital audio network solution, named mLAN (music Local Area Network), that uses IEEE 1394 as its underlying network technology. IEEE 1394 is a digital network technology that is specifically designed for real-time multimedia data transmission. The second generation of mLAN is based on the Plural Node Architecture, where the control of audio and MIDI routings between IEEE 1394 devices is split between two node types, namely an Enabler and a Transporter. The Transporter typically resides in an IEEE 1394 device and is solely responsible for transmission and reception of audio or MIDI data. The Enabler typically resides in a workstation and exposes an abstract representation of audio or MIDI plugs on each Transporter to routing control applications. The Enabler is responsible for configuring audio and MIDI routings between plugs on different Transporters. A Hardware Abstraction Layer (HAL) within the Enabler allows it to uniformly communicate with Transporters that are created by various vendors. A plug-in mechanism is used to provide this capability. When vendors create Transporters, they also create device-specific plug-ins for the Enabler. These plug-ins are created against a Transporter HAL Application Programming Interface (API) that defines methods to access the capabilities of Transporters. An Open Generic Transporter (OGT) guideline document which models all the capabilities of Transporters has been produced. These guidelines make it possible for manufacturers to create Transporters that make use of a common plug-in, although based on different hardware architectures. The introduction of the OGT concept has revealed additional Transporter capabilities that are not incorporated in the existing Transporter HAL API. This has led to the underutilisation of OGT capabilities. The main goals of this investigation have been to improve the Enabler’s plug-in mechanism, and to incorporate the additional capabilities that have been revealed by the OGT into the Transporter HAL API. We propose a new plug-in mechanism, and a new Transporter HAL API that fully utilises both the additional capabilities revealed by the OGT and the capabilities of existing Transporters.

The proliferation of PALs (Public Access Locations) is fuelling the development of new standards, protocols, services, and applications for WLANs (Wireless Local Area Networks). PALs are set up at public locations to meet continually changing, multiservice, multi-protocol user requirements. This research investigates the essential infrastructural requirements that will enable further proliferation of PALs, and consequently facilitate ubiquitous computing. Based on these requirements, an extensible architectural framework for PAL management systems that inherently facilitates the provisioning of multiple services and multiple protocols on PALs is derived. The ensuing framework, which is called Xobogel, is based on the microkernel architectural pattern, and the IPDR (Internet Protocol Data Record) specification. Xobogel takes into consideration and supports the implementation of diverse business models for PALs, in respect of distinct environmental factors. It also facilitates next-generation network service usage accounting through a simple, flexible, and extensible XML based usage record. The framework is subsequently validated for service element extensibility and simplicity through the design, implementation, and experimental deployment of SEHS (Small Extensible Hotspot System), a system based on the framework. The robustness and scalability of the framework is observed to be sufficient for SMME deployment, withstanding the stress testing experiments performed on SEHS. The range of service element and charging modules implemented confirm an acceptable level of flexibility and extensibility within the framework.

Avec l’apparition des nouvelles technologies de communication, le nombre des systèmes embarqués avionique et automobile est en constante augmentation. La gestion des communications entre ces systèmes devient alors de plus en plus complexe à mettre en oeuvre dans un contexte où les contraintes temporelles et environnementales sont très fortes et où le taux d’échanges de messages en augmentation continuelle. L’utilisation optimale des réseaux pour acheminer les données tout en respectant les contraintes temporelles imposées est essentielle du point de vue de la sûreté de fonctionnement. Historiquement, pour répondre aux problématiques d’efficacité et de sûreté, les industriels ont développé une palette de réseaux embarqués dédiés à leurs applications cibles (CAN, LIN, . . . ). Ces réseaux présentaient des débits relativement faibles à un moment où un besoin croissant en bande passante se faisait ressentir. le choix d’utiliser le concept de composants dit ‘sur étagères’ (off the shelf COTS) permettait alors de pallier à ce nouveau besoin. Dans un souci de conservation des capacités des réseaux à garantir les contraintes temporelles imposées par les systèmes embarqués temps réel, les industriels ont dû adapter ce concept de composants sur étagères aux systèmes embarqués. L’intérêt de l’utilisation de ces composants est un gain non négligeable en bande passante et en poids pour des coûts de développements relativement faibles. L’introduction de ces composants nouveaux s’est faite de telle sorte que leur impact sur les standards préexistants et les systèmes connectés soit minimal. C’est ainsi que les réseaux dit ‘hétérogènes’ ont vu leur apparition. Ces réseaux constituent une hybridation entre les technologies embarquées historiques et les composants sur étagère. Ils consistent en des réseaux d’extrémité utilisant des technologies éprouvées (telles que le CAN) interconnectés via des passerelles à un réseau fédérateur (backbone) utilisant des composants sur étagères. Dès lors, le défi majeur à relever lors de l’utilisation d’un réseau fédérateur est de respecter les contraintes temporelles des applications sollicitant les différents réseaux. L’objectif est mis à mal sur les points d’interconnexion des réseaux hétérogènes (Passerelles). Ainsi l’approche principale utilisée pour le passage d’un réseau à un autre est l’encapsulation de trames. Pour atteindre l’optimalité de performance de cette technique plusieurs paramètres sont à prendre en compte tels que le nombre de trames à encapsuler, les ordonnancements utilisés, le coût en bande passante ainsi que l’impact sur les distributions de délais (gigue). Dans l’optique de préservation des performances des réseaux, l’objet de nos travaux porte sur l’étude, la comparaison et la proposition de techniques permettant l’interconnexion de réseaux hétérogènes temps réels à la fois pour des applications à faibles et à fortes contraintes temporelles. Après un état de l’art sur les réseaux temps réel, nous spécifions différentes techniques d’interconnexion de réseaux hétérogènes, puis, nous présentons une étude de cas basée sur une architecture réseau interconnectant différents bus CAN via un réseau fédérateur sans fil Wi-Fi. L’étude que nous avons menée montre, par le biais de différentes simulations, que cette architecture réseau est une bonne candidate pour la transmission de flux à contraintes temporelles faibles. Une architecture réseau interconnectant différents bus CAN via un réseau fédérateur Ethernet commuté est ensuite considérée dans une seconde étude de cas ciblant les applications à fortes contraintes temporelles. Dans un premier temps, nous prenons en compte le cas d’un réseau fédérateur Ethernet-PQSE et, dans un second temps, le cas d’Ethernet-AVB. Cette étude nous permet de montrer l’impact des différentes techniques d’interconnexion sur les délais des flux du réseau. / With the emergence of new communication technologies, the number of avionics and automotive embedded systems is constantly increasing. The management of communications between these systems becomes increasingly complex to implement in a context where temporal and environmental constraints are very strong and where messages exchange rate is continuously increasing. The optimal use of networks to transmit data while fulfilling the imposed temporal constraints is essential from a safety point of view. Historically, in order to address safety and efficiency issues, manufacturers have developed a range of embedded networks dedicated to their target applications (CAN, LIN, . . . ). These networks have relatively low bit rates at a point of time where a growing need for bandwidth was felt. To overcome this new need, the choice of using the concept of so-called ’off-theshelf’ components (COTS) has been made. In order to preserve the networks abilities to guarantee the temporal constraints imposed by the real time embedded systems, manufacturers had to adapt the concept of off-the-shelf components to embedded systems. The benefits of using these components is a non-negligible gain in bandwidth and weight for relatively low development costs. The introduction of these new components has been made in such a way that their impact on pre-existing standards and connected systems is minimal. thereby, so-called ’heterogeneous’ networks have emerged. These networks are a hybridization of historical embedded technologies and off-the-shelf components. They consist of stub networks using proven technologies (such as CAN) interconnected via gateways to a backbone network using off-the-shelf components. Thus, the major challenge while using a heterogeneous network is to respect the temporal constraints of the applications requesting the different parts of the networks. This objective can be damaged at the interconnection points of the heterogeneous networks (Gateways). The main used approach to pass frames from one network to another is the encapsulation. To achieve the optimum performances of this technique, several parameters have to be considered such as the number of frames encapsulated, the used scheduling policy, the bandwidth cost as well as the impact on delay distributions (jitter). In order to preserve networks performances, the aim of our work is to study, compare and propose techniques ennabling the interconnection of real-time heterogeneous networks for application with both soft and hard temporal constraints. After a state of the art on real-time networks, we have specified different techniques for the interconnection of heterogeneous networks, then we have presented a case study based on a network architecture interconnecting different CAN buses via a wireless backbone network (Wi-Fi ). The study we conducted shows, through various simulations, that this network architecture is a good contender for the transmission of flows with soft temporal constraints. A network architecture interconnecting different CAN busses via a switched Ethernet backbone is considered in a second case study targeting applications with hard temporal constraints. Two different Ethernet backbone networks are taken into account. We studied first, the case of a switched Ethernet-PQSE backbone network. Then the case of a switched Ethernet-AVB backbone is considered. This study enabled us to highlight the impact of the different used interconnection techniques on network flows delays.

Made available in DSpace on 2014-12-17T14:58:20Z (GMT). No. of bitstreams: 1 BrunoCD_DISSERT.pdf: 4492853 bytes, checksum: 8fad50750c043cd4dab6a1e27f414d81 (MD5) Previous issue date: 2013-02-15 / One of the current major concerns in engineering is the development of aircrafts that have low power consumption and high performance. So, airfoils that have a high value of Lift Coefficient and a low value for the Drag Coefficient, generating a High-Efficiency airfoil are studied and designed. When the value of the Efficiency increases, the aircraft s fuel consumption decreases, thus improving its performance. Therefore, this work aims to develop a tool for designing of airfoils from desired characteristics, as Lift and Drag coefficients and the maximum Efficiency, using an algorithm based on an Artificial Neural Network (ANN). For this, it was initially collected an aerodynamic characteristics database, with a total of 300 airfoils, from the software XFoil. Then, through the software MATLAB, several network architectures were trained, between modular and hierarchical, using the Back-propagation algorithm and the Momentum rule. For data analysis, was used the technique of cross- validation, evaluating the network that has the lowest value of Root Mean Square (RMS). In this case, the best result was obtained for a hierarchical architecture with two modules and one layer of hidden neurons. The airfoils developed for that network, in the regions of lower RMS, were compared with the same airfoils imported into the software XFoil / Uma das maiores preocupa??es atuais na Engenharia ? o desenvolvimento de aeronaves que possuam baixo consumo e alto desempenho. Para isso, s?o estudados e projetados perfis aerodin?micos que tenham um valor elevado de coeficiente de sustenta??o e um valor baixo para o coeficiente de arrasto, gerando um perfil de alta efici?ncia. Quanto maior o valor da efici?ncia, menor ser? o consumo de combust?vel da aeronave, melhorando assim, o seu desempenho. Neste sentido, este trabalho objetiva desenvolver uma ferramenta para cria??o de perfis aerodin?micos a partir de caracter?sticas desejadas, como coeficiente de sustenta??o e de arrasto e efici?ncia m?xima, utilizando-se um algoritmo baseado em uma Rede Neural Artificial (RNA). Para isso, inicialmente foram coletados uma base de dados de caracter?sticas aerodin?micas de um total de 300 perfis, a partir do software XFoil. Ent?o, atrav?s de uma rotina implementada no software MATLAB, foram treinadas diversas arquiteturas de redes, entre modulares e hier?rquicas, utilizando-se o algoritmo de Retropropaga??o e a regra do Momento. Para an?lise dos resultados, foi utilizada a t?cnica de valida??o cruzada, avaliando a rede que possuiu o menor valor de Erro M?dio Quadr?tico (EMQ). Neste caso, o melhor resultado obtido foi para uma arquitetura hier?rquica com dois m?dulos e uma camada de neur?nios ocultos. Os perfis aerodin?micos desenvolvidos por essa rede, nas regi?es de menor EMQ, foram comparados aos mesmos perfis importados ao software XFoil. O presente trabalho oferece como contribui??o, em rela??o a outros trabalhos que envolvem RNA aplicada ? mec?nica dos fluidos, o desenvolvimento de perfis aerodin?micos a partir de suas caracter?sticas aerodin?micas

Orientadores: Helio Waldman, Moises Renato Nunes Ribeiro / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-12T16:44:15Z (GMT). No. of bitstreams: 1 Fardin_MarconiPereira_M.pdf: 962828 bytes, checksum: b8632419bf2a2ad176b38f5257eceada (MD5) Previous issue date: 2008 / Resumo: A infra-estrutura de transporte da internet está movendo-se para um modelo formado por roteadores de alta velocidade interconectados por um núcleo de redes ópticas inteligentes. A tecnologia WDM (Wavelength Division Multiplex) já está consolidada como meio de incrementar a capacidade de transmissão nessas redes porém, existem poucas propostas de planos de controles distribuídos capazes de lidar dinamicamente com as restrições de Qualidade de Serviço (QoS - Quality of Service) em grandes redes. Neste contexto, informações sobre os estados dos enlaces são constantemente divulgadas pela rede e o mecanismo de inundação se tornou ineficiente para lidar com a disseminação das informações. Este trabalho propõe e analisa uma estrutura sobreposta baseada em tabelas de dispersão distribuídas (DHT - Distributed Hash Table) para dispor informações em um espaço virtual. Uma rede sobreposta virtual endereçada por conteúdo (CAN - Content Addressable Network), com enlaces de longa distância para obter características de mundo pequeno, é utilizada para possibilitar o roteamento com baixa ocupação dos enlaces sem penalizar demasiadamente o atraso inserido. Avaliações analíticas ilustram o ganho de desempenho em grandes redes. / Abstract: The transport layer structure of the Internet is moving towards a model of highspeed routers interconnected by intelligent optical core networks. A consensus has been reached on wavelength division multiplexing (WDM) as the way to increase the transmission capacity. However there are few proposals for distributed control planes under dynamic Quality of Service (QoS) constraints for large networks. In this context, resources availability must be frequently distributed and updated across the network; and flooding may no longer be the best strategy for information dissemination. This work proposes and assesses an overlay structure based on distributed hash tables (DHT) for information indexing over bi-dimensional virtual coordinated spaces. By adding long-distance interconnections to Content-Addressable Networks (CAN), the resulting small-world effect can reduce the number of hops per resource search. Analytical results indicate significant reduction in management traffic over the physical layer of large networks without penalizing to much the inserted delay. / Mestrado / Telecomunicações e Telemática / Mestre em Engenharia Elétrica

Orientadores: Edmundo Roberto Mauro Madeira, Luiz Fernando Bittencourt / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-24T15:34:12Z (GMT). No. of bitstreams: 1 Meneguette_RodolfoIpolito_D.pdf: 3843404 bytes, checksum: 77fcd1e3b13f66d77af6e7537e8105fc (MD5) Previous issue date: 2013 / Resumo: Redes Ad Hoc Veiculares (VANETs) constituem uma subclasse das redes Ad Hoc móveis que provê uma conexão sem fio entre veículos bem como entre veículos e dispositivos de acostamento. Essas redes proveem aplicações que focam na segurança do condutor, em fornecer maior eficiência do tráfego dos veículos nas vias públicas e em trazer algum conforto e entretenimento para os passageiros ao longo de sua viagem. Algumas dessas aplicações precisam estar conectadas com a internet através de um ponto de acesso que encontra-se no acostamento, como uma torre celular ou uma torre WiFi. Essa conexão pode gerar uma sobrecarga de mensagens de controle e pode sofrer uma troca de ponto de acesso que impacta no desempenho da aplicação. Além da interface de conexão com o ponto de acesso, o veículo é equipado com várias outras interfaces de redes vinculadas a diferentes tecnologias, podendo tirar vantagens do uso simultâneo dessas diversas interfaces de rede, com isso, maximizando a vazão e diminuindo a latência. Entretanto, esse uso pode aumentar ainda mais a sobrecarga das mensagens de controle, tendo como consequencia o aumento do tempo de troca de um ponto de acesso a outro, afetando a vazão da rede e o seu desempenho. Para superar esses desafios e prover uma boa qualidade de serviço para a aplicação, ou seja, atender aos requisitos mínimos que as aplicações das redes veiculares necessitam, desenvolvemos e avaliamos uma arquitetura para o gerenciamento de mobilidade de fluxo baseado nas classes de aplicações das redes veiculares. A arquitetura tem como objetivo minimizar o tempo da troca de conexão do fluxo, diminuir a quantidade de perda de pacotes e diminuir o tempo de entrega das mensagens, a fim de cumprir com os requisitos mínimos das classes de aplicação veicular, bem como maximizar o desempenho das aplicações. A arquitetura foi implementada no Network Simulator versão 3 para análise do comportamento da arquitetura em diversos tipos de cenário de simulação. Além da análise da arquitetura, também foram analisados alguns dos seus módulos separadamente. O resultado dessas análises, tanto dos módulos como da arquitetura, mostrou que a arquitetura proposta apresentou um baixo tempo de entrega, com menor perda de pacotes e menor atraso / Abstract: Vehicular Ad Hoc Networks (VANETs) constitutes a subclass of mobile Ad Hoc Networks that provides a wireless connection between vehicles and between vehicles and roadside devices. These networks provide applications that focus on driver safety, traffic efficiency of vehicles on public roads, and in the comfort and entertainment to passengers throughout their journey. Some of these applications need to be connected to the Internet via an access point in the roadside, like a cell tower or WiFi tower. This connection can generate an overhead of control messages and could suffer a change of access point that can impact the performance of the application. Besides the interface connected to the access point, the vehicle is equipped with other network interfaces linked with different technologies, thus the vehicular application can take advantage of the simultaneous use of these various network interfaces, thereby maximizing throughput and reducing latency. However, this use can increase the overhead of control message, and the time of exchange of one access point to another, thus affecting the network throughput and consequently the performance of the application. To overcome these challenges and provide a good quality of service to the application, i.e., attend the minimum requirements that applications for vehicular networks need, we developed and evaluated an architecture for flow mobility management based on classes of applications of vehicular networks. The architecture aims to minimize the time of the exchange of the connection flow, decrease the amount of packet loss, and reduce the delivery time of messages in order to comply with the minimum requirements for vehicular application classes as well as to maximize the application performance. The architecture was implemented in Network Simulator version 3 to analyze the behavior of the architecture in various types of scenarios. Besides the analysis of the architecture, we also analyzed some of the modules of the architecture separately. The results of these analyses of both the architecture and its modules showed a low delivery time, lower packet loss and lower delay / Doutorado / Ciência da Computação / Doutor em Ciência da Computação

Deep neural networks with millions of parameters are at the heart of many state of the art computer vision models. However, recent works have shown that models with much smaller number of parameters can often perform just as well. A smaller model has the advantage of being faster to evaluate and easier to store - both of which are crucial for real-time and embedded applications. While prior work on compressing neural networks have looked at methods based on sparsity, quantization and factorization of neural network layers, we look at the alternate approach of pruning neurons. Training Neural Networks is often described as a kind of `black magic', as successful training requires setting the right hyper-parameter values (such as the number of neurons in a layer, depth of the network, etc ). It is often not clear what these values should be, and these decisions often end up being either ad-hoc or driven through extensive experimentation. It would be desirable to automatically set some of these hyper-parameters for the user so as to minimize trial-and-error. Combining this objective with our earlier preference for smaller models, we ask the following question - for a given task, is it possible to come up with small neural network architectures automatically? In this thesis, we propose methods to achieve the same. The work is divided into four parts. First, given a neural network, we look at the problem of identifying important and unimportant neurons. We look at this problem in a data-free setting, i.e; assuming that the data the neural network was trained on, is not available. We propose two rules for identifying wasteful neurons and show that these suffice in such a data-free setting. By removing neurons based on these rules, we are able to reduce model size without significantly affecting accuracy. Second, we propose an automated learning procedure to remove neurons during the process of training. We call this procedure ‘Architecture-Learning’, as this automatically discovers the optimal width and depth of neural networks. We empirically show that this procedure is preferable to trial-and-error based Bayesian Optimization procedures for selecting neural network architectures. Third, we connect ‘Architecture-Learning’ to a popular regularize called ‘Dropout’, and propose a novel regularized which we call ‘Generalized Dropout’. From a Bayesian viewpoint, this method corresponds to a hierarchical extension of the Dropout algorithm. Empirically, we observe that Generalized Dropout corresponds to a more flexible version of Dropout, and works in scenarios where Dropout fails. Finally, we apply our procedure for removing neurons to the problem of removing weights in a neural network, and achieve state-of-the-art results in scarifying neural networks.

M.Tech. / Service Oriented Architecture has many advantages. For example, organisations can align business with Information Technology, reuse the developed functionality, reduce development and maintain cost for applications. Organisations adopt Service Oriented Architecture with the aim of automating and integrating business processes. However, it has information security vulnerabilities that should be considered. For example, applications exchange information across the Internet, where it can be tampered with. Information security is therefore one of the crucial qualities that need to be satisfied within information systems. This dissertation addresses the issue of information security within Service Oriented Architecture applications. Some organisations rely on Service Oriented Architecture governance tools when securing information in their Service Oriented Architecture environment. However, they may purchase them without investigating whether they include information security. The aim of this dissertation is to analyse whether these tools include information security. Each tool is benchmarked against the five information security services, defined by the ISO 7498/2 document and including identification and authentication, authorisation, confidentiality, integrity and non-repudiation. The dissertation concludes with a table summarising the results. This dissertation offers decision-makers information that can assist them in analysing whether Service Oriented Architecture governance tools includes information security. It also assists organisations to be aware of security vulnerabilities within Service Oriented Architecture applications, and the consequences that may arise if information security measures are ignored.