Les réseaux complexes sont des ensembles constitués d’un grand nombre d’entités interconnectées par des liens. Ils sont modélisés par des graphes dans lesquels les noeuds représentent les entités et les arêtes entre les noeuds représentent les liens entre ces entités. Ces graphes se caractérisent par un très grand nombre de sommets et une très faible densité de liens. Les réseaux sociaux sont des exemples de réseaux complexes où les entités sont des individus et les liens sont les relations (d’amitié, d’échange de messages) entre ces individus.L’analyse des réseaux complexes est généralement basée sur l’exploration locale du graphe sous-jacent : après avoir traité un nœud u, les prochains noeuds auxquels l’application fait référence appartiennent au voisinage de u. Étant donné que le graphe sous-jacent est habituellement non structuré, les séquences d’accès aux données en mémoire tendent à avoir une faible localité lorsque qu’on utilise par exemple le stockage de Yale qui est l’un des meilleurs connus. En plus, dans les applications basées sur l’analyse des réseaux le nombre de calculs requis pour chaque noeud peut être très variable, ce qui, dans les mises en œuvre parallèles (multithreadées), se traduit par un déséquilibre de charges entre les threads.Le travail réalisé dans cette thèse était lié au développement d’applications d’analyse des réseaux sociaux, qui soient à la fois faciles à écrire et efficaces. A cet effet, deux pistes ont été explorées: a)L’exploitation de la structure en communautés pour définir des techniques de stockage qui réduisent les défauts de cache lors de l’analyse des réseaux sociaux; b)La prise en compte de l’hétérogénéité des degrés des noeuds pour optimiser la mise en oeuvre parallèle.La première contribution de cette thèse met en évidence l'exploitation de la structure en communautés des réseaux complexes pour la conception des algorithmes de numérotation des graphes (NumBaCo, CN-order) permettant la réduction des défauts de cache des applications tournant dans ces graphes.Les résultats expérimentaux en mode séquentiel sur plusieurs architectures (comme Numa4) ont montré que les défauts de cache et ensuite le temps d'exécution étaient effectivement réduits; et que CN-order se sert bien des avantages des autres heuristiques de numérotation (Gorder, Rabbit, NumBaCo) pour produire les meilleurs résultats.La deuxième contribution de cette thèse a considéré le cas des applications multi-threadées. Dans ce cas, la réduction des défauts de cache n'est pas suffisante pour assurer la diminution du temps d'exécution; l'équilibre des charges entre les threads doit être assuré pour éviter que certains threads prennent du retard et ralentissent ainsi toute l'application. Dans ce sens, nous nous sommes servis de la propriéte de l'hétérogénéité des dégrés des noeuds pour développer l'heuristique Deg-scheduling. Les résultats expérimentaux avec plusieurs threads sur l'architecture Numa4 montrent que Deg-scheduling combiné aux heuristiques de numérotation permet d'obtenir de meilleur résultats.La dernière contribution de cette thèse porte sur l'intégration des deux catégories d'heuristiques développées dans les DSLs parallèles d'analyse des graphes. Par exemple, avec le DSL Green-Marl, les performances sont améliorées à la fois grâce aux heuristiques de numérotation et grâce aux heuristiques d’ordonnancement (temps réduit de 35% grâce aux heuristiques). Mais avec le DSL Galois, les performances sont améliorées uniquement grâce aux heuristiques de numérotation (réduction de 48%). / A complex network is a set of entities in a relationship, modeled by a graph where nodes represent entities and edges between nodes represent relationships. Graph algorithms have inherent characteristics, including data-driven computations and poor locality. These characteristics expose graph algorithms to several challenges, because most well studied (parallel) abstractions and implementation are not suitable for them. The main question in this thesis is how to develop graph analysis applications that are both --easy to write (implementation challenge), -- and efficient (performance challenge)? We answer this question with parallelism (parallel DSLs) and also with knowledge that we have on complex networks (complex networks properties such as community structure and heterogeneity of node degree).The first contribution of this thesis shows the exploitation of community structure in order to design community-aware graph ordering for cache misses reduction. We proposed NumBaCo and compared it with Gorder and Rabbit (which appeared in the literature at the same period NumBaCo was proposed). This comparison allowed to design Cn-order, another heuristic that combines advantages of the three algorithms (Gorder, Rabbit and NumBaCo) to solve the problem of complex-network ordering for cache misses reduction. Experimental results with one thread on Core2, Numa4 and Numa24 (with Pagerank and livejournal for example) showed that Cn-order uses well the advantages of the other orders and outperforms them.The second contribution of this thesis considered the case of multiple threads applications. In that case, cache misses reduction was not sufficient to ensure execution time reduction; one should also take into account load balancing among threads. In that way, heterogeneity of node degree was used in order to design Deg-scheduling, a heuristic to solve degree-aware scheduling problem. Deg-scheduling was combined to Cn-order, NumBaCo, Rabbit, and Gorder to form respectively Comm-deg-scheduling, Numb-deg-scheduling, Rab-deg-scheduling and Gor-deg-scheduling. Experimental results with many threads on Numa4 showed that Degree-aware scheduling heuristics (Comm-deg-scheduling, Numb-deg-scheduling, Rab-deg-scheduling and Gor-deg-scheduling) outperform their homologous graph ordering heuristics (Cn-order, NumBaCo, Rabbit, and Gorder) when they are compared two by two.The last contribution was the integration of graph ordering heuristics and degree-aware scheduling heuristics in graph DSLs and particularly Galois and Green-Marl DSLs. We showed that with Green-Marl, performances are increased by both graph ordering heuristics and degree-aware scheduling heuristics (time was reduced by 35% due to heuristics). But with Galois, performances are increased only with graph ordering heuristics (time was reduced by 48% due to heuristics).In perspective, instead of using complex networks properties to design heuristics, one can imagine to use machine learning. Another perspective concerns the theoretical aspect of this thesis. We showed that graph ordering for cache misses reduction and degree-aware scheduling for load balancing problems are NP-complete. We provided heuristics to solve them. But we didn't show how far these heuristics are to the optimal solutions. It is good to know it in the future.

Orientador: Yuzo Iano / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-20T16:13:03Z (GMT). No. of bitstreams: 1 Noboa_FranciscoJoseViudes_M.pdf: 2562716 bytes, checksum: 8090bd81ec54aaeab629033ee45a1ef0 (MD5) Previous issue date: 2012 / Resumo: Considerando-se que para controlar e possibilitar a entrega de diversos conteúdos e serviços a qualquer tipo de acesso fazia-se necessário um núcleo bem definido e estruturado, surgiu o IMS (IP Multimedia Subsystem) com o propósito de prover a integração completa das redes e serviços. O IMS define uma arquitetura completa e framework que habilita a convergência de voz, vídeo, dados e tecnologia de rede móvel através de uma infraestrutura baseada em IP, preenchendo a distância entre os dois paradigmas de comunicação mais bem sucedidos, celular e tecnologia IP. Este trabalho apresenta a arquitetura IMS como controle central de todas as redes, e uma arquitetura para o desenvolvimento de aplicações móveis que incorporem voz, vídeo e dados. A arquitetura IMS apesar de promissora apresenta inúmeras oportunidades de melhoria no seu mecanismo de segurança, muitas dessas oportunidades de melhoria estão relacionadas a falhas e até mesmo a falta de especificações de segurança quando a arquitetura foi originalmente implementada. Assim o maior desafio na implantação e globalização da arquitetura IMS são as falhas de segurança e a vulnerabilidade que a arquitetura possui a diversos tipos de ataques que podem atingir e prejudicar, tanto operadoras quanto usuários da rede. O objetivo deste projeto de pesquisa é apresentar uma análise detalhada sobre a estrutura da arquitetura IMS focando principalmente na arquitetura de segurança desenvolvida pela 3GPP, e assim prover análises e soluções para os ataques, vulnerabilidades e falhas de segurança que atingem a arquitetura tanto do ponto de vista de operadoras e provedores de serviço quanto na perspectiva do usuário / Abstract: Considering that to control and enable the delivery of diverse content and services to any type of access it should make necessary a core well-defined and structuralized, appeared the IMS (IP Multimedia Subsystem) in order to provide the complete integration of networks and services. IMS defines a complete architecture and framework that enables the convergence of voice, video, data and mobile network technology over an infrastructure based on IP, in addition to filling the gap between the two communications paradigms most successful, cellular and IP technology. This paper presents the IMS architecture as central control of all networks, and architecture for developing mobile applications that incorporate voice, video and data. The IMS architecture despite being promising presents numerous opportunities for improvement in its security mechanism, many of these opportunities for improvement are related to failures and even the lack of security specifications when creating the architecture. So the biggest challenge in the implementation of the IMS architecture and globalization are the security issues and vulnerabilities that the architecture faces being vulnerable to several types of attacks that can reach and affect carriers and network users. The objective of this research project is to present a detailed analysis on the structure of the IMS architecture focusing primarily on security architecture developed by 3GPP, and thus provide analysis and solutions about the attacks, vulnerabilities and security issues that affect the architecture from the point of view of carriers, service providers and from the network users / Mestrado / Telecomunicações e Telemática / Mestre em Engenharia Elétrica

The focus of this dissertation is on large-scale and long-running networked real-time multiplayer games. In this type of games, each player controls one or many entities, which interact in a shared virtual environment. Three attributes - scalability, security, and fault tolerance - are considered essential for this type of games. The normal approaches for building this type of games, using a client/server or peer-to-peer architecture, fail in achieving all three attributes. We propose a server-network architecture that supports these attributes. In this architecture, a cluster of servers collectively manage the game state and each server manages a separate region of the virtual environment. We discuss how the architecture can be extended using proxies, and we compare it to other similar architectures. Further, we investigate how a distributed database management system can support the proposed architecture. Since efficiency is very important in this type of games, some properties of traditional database systems must be relaxed. We also show how methods for increasing scalability, such as interest management and dead reckoning, can be implemented in a database system. Finally, we suggest how the proposed architecture can be validated using a simulation of a large-scale game.

The Named Data Networking (NDN) is a new Internet architecture that changes the network semantic from packet delivery to content retrieval and promises benefits in areas such as content distribution, security, mobility support, and application development. While the basic NDN architecture applies to any network environment, local area networks (LANs) are of particular interest because of their prevalence on the Internet and the relatively low barrier to deployment. In this dissertation, I design NDN protocols and implement NDN software, to make NDN communication in LAN robust and efficient. My contributions include: (a) a forwarding behavior specification required on every NDN node; (b) a secure and efficient self-learning strategy for switched Ethernet, which discovers available contents via occasional flooding, so that the network can operate without manual configuration, and does not require a routing protocol or a centralized controller; (c) NDN-NIC, a network interface card that performs name-based packet filtering, to reduce CPU overhead and power consumption of the main system during broadcast communication on shared media; (d) the NDN Link Protocol (NDNLP), which allows the forwarding plane to add hop-by-hop headers, and provides a fragmentation-reassembly feature so that large NDN packets can be sent directly over Ethernet with limited MTU.

Orientador: Christian Rodolfo Esteve Rothenberg / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-27T08:39:20Z (GMT). No. of bitstreams: 1 Siqueira_MarcosAntoniode_D.pdf: 7263131 bytes, checksum: fad0f39a012c338503de664fbec8fe8c (MD5) Previous issue date: 2015 / Resumo: Esta tese apresenta uma proposta de arquitetura para controle de redes ópticas de transporte que utiliza o paradigma de redes definidas por software, com suporte a operação autonômica com base em políticas. A arquitetura é constituída pelos seguintes pilares: (i) modelagem dos elementos de rede, incluindo suas interconexões, restrições, capacidades, entre outros, utilizando a linguagem YANG; (ii) composição dos modelos dos elementos de rede e suas relações em um modelo que representa a rede, suportando transformações para representação da rede como grafos de propriedades; e (iii) um modelo de políticas baseado em objetos associados ao grafo de propriedades da rede que viabiliza a operação autonômica do controlador. A proposta foi validada através de provas de conceito realizadas por simulações, protótipos e experimentos, incluindo casos de uso de segmentação e virtualização da rede óptica de transporte, aplicações SDN para ajuste de parâmetros operacionais da rede com base em políticas, bem como a operação autônoma do controlador SDN com auxílio de ferramentas de simulação com rotinas de planejamento automatizado / Abstract: This thesis proposes an architecture for optical transport networks control, using the software defined networking paradigm, with support for policy-based autonomic operation. The architecture is composed of three pillars: (i) modeling of network elements, its interconnections, constraints and capabilities using the YANG language; (ii) composition of the network element models and its interconnections forming a network model, supporting transformations for representing the network as property graphs; and (iii) a policy model based on objects associated to the network graph designed for allowing autonomic operation of the network controller. The proposal has been validated through a set of proofs of concept performed via simulations, prototypes and experiments, including use cases for optical transport network slicing and virtualization, SDN applications for policy-based operational parameters adjustment, and autonomic operation of the SDN controller assisted by simulation tools with routines for automated planning / Doutorado / Engenharia de Computação / Doutor em Engenharia Elétrica

Network virtualization is a technology that enables multiple virtual instances to coexist on a common physical network infrastructure. This paradigm fostered new business models, allowing infrastructure providers to lease or share their physical resources. Each virtual network is isolated and can be customized to support a new class of customers and applications. To this end, infrastructure providers need to embed virtual networks on their infrastructure. The virtual network embedding is the (NP-hard) problem of matching constrained virtual networks onto a physical network. Heuristics to solve the embedding problem have exploited several policies under different settings. For example, centralized solutions have been devised for small enterprise physical networks, while distributed solutions have been proposed over larger federated wide-area networks. In this thesis we present a policy-based architecture for the virtual network embedding problem. By policy, we mean a variant aspect of any of the three (invariant) embedding mechanisms: physical resource discovery, virtual network mapping, and allocation on the physical infrastructure. Our architecture adapts to different scenarios by instantiating appropriate policies, and has bounds on embedding efficiency, and on convergence embedding time, over a single provider, or across multiple federated providers. The performance of representative novel and existing policy configurations are compared via extensive simulations, and over a prototype implementation. We also present an object model as a foundation for a protocol specification, and we release a testbed to enable users to test their own embedding policies, and to run applications within their virtual networks. The testbed uses a Linux system architecture to reserve virtual node and link capacities.

In der heutigen, digitalisierten Gesellschaft ist es für Internet Service Provider (ISP) zunehmend wichtig, den ansteigenden Anforderungen ihrer Endnutzer gerecht zu werden. Eine Möglichkeit diese Herausforderung zu adressieren ist es, die vorhandene Netzwerkinfrastruktur durch den Einsatz von Traffic Engineering Technologien möglichst effizient zu betreiben. Segment Routing (SR) bietet eine elegante Möglichkeit Traffic Engineering einzusetzen ohne viel Overhead zu erzeugen. Obwohl es bereits verschiedene Ansätze zum SR Traffic Engineering gibt, verbleiben die meisten dieser Arbeiten auf einer sehr theoretischen Ebene und versäumen es, praktische Anforderungen und Nebenbedingungen zu erkennen und miteinzubeziehen. Diese Dissertation schließt die Lücke zwischen Theorie und Praxis, indem die spezifischen Anforderungen eines Tier 1 ISPs identifiziert werden, neue Optimierungsmodelle kreiert werden, welche ebendiese Nebenbedingungen erfüllen, und, schlussendlich, eine detaillierte Evaluation mittels realer Messdaten von selbigem ISP durchgeführt wird. Die Optimierungsmodelle berücksichtigen praktisch relevante Anforderungen wie die Minimierung der Anzahl der verwendeten SR Tunnel und SR Label, die proaktive Minimierung von Überlast in einer Vielzahl verschiedener Fehlerszenarien und weitere. Es wird darüber hinaus gezeigt, dass SR Optimierungen größtenteils unabhängig von der zugrundeliegenden Linkmetrikkonfiguration funktioniert und dass SR das Potential aufweist, Routingkonfigurationen zu erzeugen, die für längere Zeitabschnitte stabil bleiben.

Indiana University-Purdue University Indianapolis (IUPUI) / This thesis firstly surveys and summarizes the state-of-the-art studies from two research areas in Software De fined Networking (SDN) architecture: (i) control plane scalability and (ii) Quality of Service (QoS)-related problems. It also outlines the potential challenges and open problems that need to be addressed further for more scalable SDN control planes and better and complete QoS abilities in SDN networks. The thesis secondly presents a hierarchical SDN design along with an inter-AS QoS-guaranteed routing approach. This design addresses the scalability problems of control plane and privacy concerns of inter-AS QoS routing philosophies in SDN. After exploring the roots of control plane scalability problems in SDN, the thesis then proposes a metric to quantitatively evaluate the control plane scalability in SDN. Later, the thesis presents a general framework for economic analysis of network architectures and designs. To this end, the thesis defines and utilizes two metrics, Unit Service Cost Scalability and Cost-to-Service, to evaluate how SDN architecture performs compared to MPLS architecture in terms of unit cost for a service and cost of introducing a new service along with giving mathematical models to calculate Capital Expenditures (CAPEX) and Operational Expenditures (OPEX) of a network. Moreover, the thesis studies the problem of optimal final pricing for services by proposing an optimal pricing scheme for a service request with QoS in SDN environment while aiming to maximize benefits of both service providers and customers. Finally, the thesis investigates how programmable network architectures, i.e. SDN, affect the network economics compared to traditional network architectures, i.e. MPLS, in case of failures along with exploring the economic impact of failures in different SDN control plane models.

No / The purpose of this study was to determine whether artificial neural network (ANN) programs implementing different backpropagation algorithms and default settings are capable of generating equivalent highly predictive models. Three ANN packages were used: INForm, CAD/Chem and MATLAB. Twenty variants of gradient descent, conjugate gradient, quasi-Newton and Bayesian regularisation algorithms were used to train networks containing a single hidden layer of 3¿12 nodes. All INForm and CAD/Chem models trained satisfactorily for tensile strength, disintegration time and percentage dissolution at 15, 30, 45 and 60 min. Similarly, acceptable training was obtained for MATLAB models using Bayesian regularisation. Training of MATLAB models with other algorithms was erratic. This effect was attributed to a tendency for the MATLAB implementation of the algorithms to attenuate training in local minima of the error surface. Predictive models for tablet capping and friability could not be generated. The most predictive models from each ANN package varied with respect to the optimum network architecture and training algorithm. No significant differences were found in the predictive ability of these models. It is concluded that comparable models are obtainable from different ANN programs provided that both the network architecture and training algorithm are optimised. A broad strategy for optimisation of the predictive ability of an ANN model is proposed.

Transformer neural networks have made a significant impact on natural language processing. The Transformer network self-attention mechanism effectively addresses the vanishing gradient problem that limits a network learning capability, especially when the time series gets longer or the size of the network gets deeper. This dissertation examines the usage of the Transformer model for time-series forecasting and customizes it for a simultaneous multistep-ahead prediction model in a surrogate model predictive control (MPC) application. The proposed method demonstrates enhanced control performance and computation efficiency compared to the Long-short term memory (LSTM)-based MPC and one-step-ahead prediction model structures for both LSTM and Transformer networks. In addition to the Transformer, this research investigates hybrid machine-learning modeling. The machine learning models are known for superior function approximation capability with sufficient data. However, the quantity and quality of data to ensure the prediction precision are usually not readily available. The physics-informed neural network (PINN) is a type of hybrid modeling method using dynamic physics-based equations in training a standard machine learning model as a form of multi-objective optimization. The PINN approach with the state-of-the-art time-series neural networks Transformer is studied in this research providing the standard procedure to develop the Physics-Informed Transformer (PIT) and validating with various case studies. This research also investigates the benefit of nonlinear model-based control and estimation algorithms for managed pressure drilling (MPD). This work presents a new real-time high-fidelity flow model (RT-HFM) for bottom-hole pressure (BHP) regulation in MPD operations. Lastly, this paper presents details of an Arduino microcontroller temperature control lab as a benchmark for modeling and control methods. Standard benchmarks are essential for comparing competing models and control methods, especially when a new method is proposed. A physical benchmark considers real process characteristics such as the requirement to meet a cycle time, discrete sampling intervals, communication overhead with the process, and model mismatch. Novel contributions of this work are (1) a new MPC system built upon a Transformer time-series architecture, (2) a training method for time-series machine learning models that enables multistep-ahead prediction, (3) verification of Transformer MPC solution time performance improvement (15 times) over LSTM networks, (4) physics-informed machine learning to improve extrapolation potential, and (5) two case studies that demonstrate hybrid modeling and benchmark performance criteria.