This master's thesis deals with security of wireless network standard 802.11. There are described security mechanisms as hide SSID, MAC address filtering, WEP, WPA, WPA2 protocols. There are described also the most often kinds of attacks (Denial of Service, disclosure WEP key, Man-in-the-Middle). The work contains investigation of used security mechanisms in wireless networks in parts of Brno. There were used NetStumbler and inSSIDer applications for this measurement. The last part of work contains measurement of security mechanisms (WEP, WPA, WPA2) effect to transfer rate from the point of view of the end user. The measurement was realized with Iperf application.

This diploma thesis deals with the topic of proxy servers and firewalls and considers other associated technologies and network techniques. It systematically describes the general issues of firewalls, with a special focus on proxy firewalls and their safety. Additional systems mentioned in this document are intrusion detection systems (IDS), antivirus systems and content control filters – as these are also connected with safety of networks, servers and workstations or with limiting various Internet sources. IDS systems can be typically supplemented with various additional applications or tools that enrich them and increase their potential – including graphic additions. This part is remembered too. Some systems can communicate with each other, which is successfully utilised (FW & IDS co-operation, for example). The purpose of the first large chapter is to present firewall technologies, to list firewall types, their basic functionality and to present the final comparison. It marginally mentions firewall applications in practice. Chapter two explains the theory of network address translation (NAT), deals with its functionality, safety and with limiting the NAT mechanism. Chapter three brings a comprehensive presentation of proxy servers. It explains their principle from the point of view of functionality and the specification of application areas. The chapter is complete with a clear list of proxy server types and their descriptions. The last chapter named Linux Proxy Firewall is the key part of the work. It deals generally with the Linux platform, the Debian GNU/Linux distribution, principles of safety policy, network configuration, network server safety, Linux firewalls (Netfilter framework, Iptables tool) and with the Squid proxy server. The following subchapters respect the previous structure: they describe the theories of intrusion detection systems, antivirus checks and content filtering based on different methods. All this is presented similarly to the previous chapters. A proxy firewall solution built on the Linux operating system has been proposed in the practical part. The Debian GNU/Linux distribution has been chosen, being very suitable for server use due to its features. This environment is also used for additional safety software contained in the proxy firewall: antivirus protection, content filtering and an intrusion detection system. The priority is the most comprehensive computer network security, which requires detection abilities with the broadest possible coverage in the area of network safety. The purpose of this diploma thesis is not only to describe the principle of operation of proxy servers and to compare them with other types and other systems, but it also brings my own proposed free solution, which increases network safety and has the ambition of comparing it with clearly commercial products available on the market.

Network attack analysis using honeyd tool. Opensource honeypots WinHoneyd and LaBrea deployment testing. Description and solving deployment problems conected with applied operating system. Capture of network attack with packet sniffer. Captured data analyzing and procesing

This work is focused on the security of small to medium-sized networks with central administration, especially on the creation of a methodology for secure network design.The design of a secure network for a start-up IT company, using open-source software, is described. Deployment of the designed secure network, with central management, is performed and the connectivity of network elements are tested. The model simulates network traffic situations and network attacks using penetration testing techniques. In terms of the severity of the impact on network security, intercepted attacks are evaluated and immediately reported to responsible persons. Finally, the results of the intercepted attacks are processed and further actions are recommended.

This thesis is focused on the study of the best-known network attacks, especially on those that can be theoretically detected without knowledge of the contents of transmitted messages. The goal is to use the basis of acquired knowledge to create a tool that will simulate the behavior of the communication in different network attacks. Simulation outputs will be used for testing the quality of security tools designed to defend against network attacks. The simulator will be used only for offline testing, it will not be possible to carry out real attacks. Purpose of this work is to improve the security against network attacks nowadays.

Nesta tese investigam-se três pontos ligados a fragilidades topológicas de grafos e suas aplicações a redes complexas reais e, em especial, a redes de relacionamentos criminais. Na primeira etapa, apresenta-se in abstracto um método inédito e eficiente de fragmentação de redes complexas por módulos. O procedimento identifica em primeiro lugar comunidades topológicas por meio da qual a rede pode ser representada usando algoritmos heurísticos de extração de comunidades. Então, somente os nós que participam de ligaçõees inter-comunitaárias são removidos em ordem decrescente de sua centralidade de intermediação. Ilustra-se o método pela aplicação a uma variedade de redes reais nas áreas social, de infraestrutura, e biológica. Mostra-se que a abordagem por módulos supera ataques direcionados a vértices baseados somente no ordenamento de índices de centralidade, com ganhos de eficiência fortemente relacionados à modularidade da rede.No segundo momento, introduzem-se os conceitos de robustez e fragilidade de redes generalizadas para avaliar o quanto um determinado sistema se comporta frente a ataques incompletos. Ainda, avalia-se o desempenho (relação entre robustez e custo computacional) de diversos ataques sequenciais e simultâneos a redes modulares por meio de uma medida empírica que chamamos de performance. Mostra-se por meio de redes artificiais de referência e de redes reais que para sistemas altamente modulares a estratégia de fragmentação por módulos apresenta um desempenho até 10 vezes superior aos demais ataques. Na última etapa, explora-se com maior profundidade a natureza subjacente de redes reais de relacionamentos criminais. Apresenta-se uma rede única e sem precedentes construída pela Polícia Federal Brasileira consistindo de mais de 35.000 relacionamentos entre 24.000 indivíduos. Os dados foram coletados entre abril e agosto de 2013 e consistem em informações fornecidas diretamente pelos investigadores responsáveis de cada caso. O sistema apresenta características típicas de redes sociais, porém é bem mais “escuro"que o comportamento típico, com baixos níveis tanto de densidade de arestas quanto de eficiência de rede. Além do mais, o sistema é extremamente modular o que implica ser possível desmantelar toda a rede de crimes federais brasileiros com a remoção de aproximadamente 2% dos indivíduos escolhidos conforme a prescrição do método modular. Também, a rede é controlável no sentido da teoria matemática de controle, significando que com acesso a aproximadamente 20% dos nós é possível, em tese, levar qualquer variável dinâmica de um estado inicial a um estado final arbitrário em um tempo finito. Exibi-se tambám uma análise topológica e de fragilidades de uma segunda rede criminal relacionada a investigações da Polícia Federal. Trata-se de um fórum online destinado à prática de crimes cibernéticos na chamada camada profunda da internet (deep web). (Continuação ) Após a coleta dos dados foi possível construir uma rede de relacionamentos com quase 10.000 indivíduos. Comparou-se, entãoo, a estratégia usada de fato pela Polícia Federal durante a Operação Darknet com a previsão teórica de ataques topológicos à rede criminal e mostrou-se que ataques dirigidos por grau teriam fragmentado o sistema de maneira quase 15 vezes mais eficiente. Por outro lado, esta rede não é modular apesar de novamente apresentar uma arquitetura mais “escura" que o usual. Por termo, demonstra-se que os ataques por arestas estão diretamente relacionados ao aprisionamento enquanto que a ressocialização e/ou morte dos indivíduos é melhor interpretada como a remoção por vértices. Destarte, comprovou-se que de um ponto de vista topológico a ressocialização é de fato mais eficiente em reduzir a criminalidade do que o aprisionamento. Contudo, na rede de crimes federais estudada essa diferenca é muito pequena, de tal modo que ambas as políticas poderiam, em tese, ser aplicadas a fim de se combater eficientemente o sistema criminoso. / In this thesis we investigate three points connected to topological fragilities of graphs and their applications to real complex networks and, in particular, to networks of criminal relationships. In the first step, we present an unprecedented and efficient method of fragmentation of complex networks by modules. Firstly, the procedure identifies topological communities through which the network can be represented using heuristic communities extraction algorithms. After that, only the nodes that bridge communities are removed in descending order of their betweenness centrality . We illustrate the method by the applying it to a variety of real networks in the social, infrastructure, and biological fields. We show that the modular approach outperforms attacks traditional attacks based only on the ordering of centrality indexes, with efficiency gains strongly related to the modularity of the network. In the second moment, we introduce the concepts of generalized robustness and fragility of networks to evaluate how much a certain system behaves in the face of incomplete attacks. Also, we evaluate the relation between robustness and computational cost of several sequential and simultaneous attacks to modular networks by means of an empirical measure that we call performance. In this sense, we show through artificial and real networks that for highly modular systems the strategy of fragmentation by modules presents a performance up to 10 times superior to traditional attacks. In the last step, we explore in more depth the underlying nature of real networks of criminal relationships. We present a unique and unprecedented network built by the Brazilian Federal Police consisting of more than 35,000 relationships among 24,000 individuals. The data were collected between April and August 2013 and consist of information provided directly by the investigators responsible for each case. The system has typical characteristics of social networks, but is much "darker"than traditional social networks, with low levels of edge density and network efficiency. Moreover, the network is extremely modular which implies that it is possible to dismantle all the network of Brazilian federal crimes with the removal of approximately 2% of the individuals chosen according to the modular method. Also the network is controllable in the sense of the mathematical control theory, meaning that with access only to 20% of nodes it is possible, In theory, to take any dynamic variable from an initial state to an arbitrary final state in a finite time. We also show a topological analysis of a second criminal network related to Federal Police investigations. This is an online forum for cybercrime in the so-called deep web. After the data collection, it was possible to build a network of relationships with almost 10,000 individuals. We then compared the strategy actually used by the Federal Police during Operation Darknet with the theoretical prediction of topological attacks on the criminal network and showed that degree-based attacks would have fragmented the system almost 15 times more efficiently. On the other hand, this network is not modular despite presenting a "darker"architecture than usual. As a last result, this particular system is not controllable in practical terms. We finish the study by showing that edge attacks are directly related to the imprisonment whereas the resocialization and/or death of the individuals is better interpreted as the removal of vertices. Thus, we prove that from a topological point of view resocialization is in fact more efficient in reducing crime rates than imprisonment. However, in the network of federal crimes studied here this difference is very small, so that both policies could in theory be applied in order to combat effectively the criminal system.

Ένα φαινόμενο που έκανε την εμφάνισή του τα τελευταία χρόνια είναι η μελέτη μεγάλων δικτύων που εμφανίζουν μια ιεραρχική δομή ανεξαρτήτως κλίμακας (large scale-free networks). Μια παραδοσιακή μέθοδος μοντελοποίησης δικτύων είναι η χρήση γραφημάτων και η χρησιμοποίηση αποτελεσμάτων που προκύπτουν από την Θεωρία Γράφων. Όμως στα κλασικά μοντέλα που έχουν μελετηθεί, δυο κόμβοι του ίδιου γραφήματος έχουν την ίδια πιθανότητα να συνδέονται με οποιουσδήποτε δυο άλλους κόμβους. Αυτός ο τρόπος μοντελοποίησης αποτυγχάνει να περιγράψει πολλά δίκτυα της καθημερινής ζωής, όπως δίκτυα γνωριμιών όπου οι κόμβοι συμβολίζουν ανθρώπους και συνδέονται μεταξύ τους αν γνωρίζονται άμεσα. Σε ένα τέτοιο δίκτυο είναι αναμενόμενο δυο φίλοι κάποιου ατόμου να έχουν μεγαλύτερη πιθανότητα να γνωρίζονται μεταξύ τους από ότι δυο τυχαία επιλεγμένοι ξένοι. Αυτό ακριβώς το φαινόμενο ονομάζεται συσσωμάτωση (clustering) και είναι χαρακτηριστικό για τα εν λόγω δίκτυα. Είναι γεγονός ότι πολλά δίκτυα που συναντώνται στη φύση αλλά και πάρα πολλά ανθρωπογενή δίκτυα εντάσσονται σε αυτήν την κατηγορία. Παραδείγματα τέτοιων είναι τα δίκτυα πρωτεϊνών, δίκτυα τροφικών αλυσίδων, επιδημικής διάδοσης ασθενειών, δίκτυα ηλεκτρικού ρεύματος, υπολογιστών, ιστοσελίδων του Παγκόσμιου Ιστού, δίκτυα γνωριμιών, επιστημονικών αναφορών (citations) κ.α. . Παρότι φαίνεται να άπτονται πολλών επιστημών όπως η Φυσική, η Βιολογία, η Κοινωνιολογία και η Πληροφορική, δεν έχουν τύχει ευρείας μελέτης, καθώς μέχρι στιγμής έλειπαν πραγματικά μεγάλα δίκτυα για πειραματική μελέτη (κενό που καλύφθηκε με την ανάπτυξη του Παγκόσμιου Ιστού). Μέχρι σήμερα δεν έχουν φωτιστεί όλα εκείνα τα σημεία και τα μεγέθη που είναι χαρακτηριστικά για αυτά τα δίκτυα και που πρέπει να εστιάσει η επιστημονική έρευνα, παρόλα αυτά έχει γίνει κάποια πρόοδος. Μια τέτοια έννοια που μπορεί να εκφραστεί με πολλά μεγέθη είναι η έννοια της κεντρικότητας (centrality) ενός κόμβου στο δίκτυο. Η χρησιμότητα ενός τέτοιου μεγέθους, αν μπορεί να οριστεί, είναι προφανής, για παράδειγμα στον τομέα της εσκεμμένης «επίθεσης» σε ένα τέτοιο δίκτυο (π.χ. δίκτυο υπολογιστών). Η ακριβής όμως συσχέτιση της κεντρικότητας με τα άλλα χαρακτηριστικά μεγέθη του δικτύου, όπως η συσσωμάτωση, δεν είναι γνωστή. Στόχος της εργασίας είναι να εμβαθύνει στην έννοια της κεντρικότητας, και χρησιμοποιεί σαν πεδίο πειραματισμών τον χώρο της εσκεμμένης επίθεσης σε ανεξάρτητα κλίμακας δίκτυα. Στο πλαίσιο αυτό γίνεται μια συνοπτική παρουσίαση των μοντέλων δικτύων που έχουν προταθεί μέχρι σήμερα και αναλύεται η έννοια της κεντρικότητας μέσω των παραδοσιακών ορισμών της από την επιστήμη της Κοινωνιολογίας. Στη συνέχεια προτείνεται μια σειρά ορισμών της κεντρικότητας που την συνδέουν με μεγέθη του δικτύου όπως ο συντελεστής συσσωμάτωσης. Η καταλληλότητα των ορισμών αυτών διαπιστώνεται στην πράξη, εξομοιώνοντας πειραματικά επιθέσεις σε ανεξάρτητα κλίμακας μεγάλα δίκτυα και χρησιμοποιώντας στρατηγικές επίθεσης που βασίζονται σε αυτές. / A trend in recent years is the study of large networks which possess a hierarchical structure independent of the current scale (large scale-free networks). A traditional method of network modelling is the use of graphs and the usage of results based on Graph Theory. Until recently, the classical models studied, describe the probability of two random vertices connecting with each other as equal for all pairs of vertices. This modelling fails to describe many everyday networks such as acquaintance networks, where the vertices are individuals and connect with an edge if they know each other

Cyber has been around for over a decade and yet we are still faces with a situation of a very weak or rather no regulation. This is being heavily influenced by our weak technological development and due to the nature of cyberspace and the Internet. Cyber warfare poses interesting questions for us. It is one of a kind type of warfare, the one we cannot see or feel, but it‘s impacts are instantaneous and potentially devastating. States and scholars agree on that. This thesis attempts to explore possibilities of application of existing international laws to amend this situation and answer the questions if it is adequate or at all possible. While scholars are still arguing about the basics of what cyber warfare and cyberspace actually is, it keeps on evolving. States on the other hand have recognized the potential threat of cyber warfare a long time ago and are attempting to mend the existing legal void, however not successfully. The effects of their efforts are limited only to a small number of States. States who are not willing to give up their cyber capability would also stay clear from such international legislation. International treaties and State practice were analyzed in search of a way to accommodate cyber warfare under the current regime. The findings show that application of existing legal basis to cyber warfare is at best difficult and strained. The reality is that cyber warfare does not fit adequately under any of the legal umbrellas at the moment. Application of... [to full text] / Kibernetinis karas jau egzistuoja daugiau nei dešimtmeti tačiau mes vis dar turime labai silpną šio reiškinio reguliavimą. Tokia situacija yra stipriai įtakota mūsų silpnu techniniu galimybių bei interneto struktūros. Kibernetinis karas yra labai keblus. Tai yra naujoviškas kariavimo būdas kurio mes nematome, bet jo pasekmes gali būti žaibiškos ir niokojančios. Mokslininkai ir pasaulio valstybės tai jau seniai pripažino. Šis darbas bando atskleisti galimybes tarptautines teises reguliavimui kibernetinio karo atžvilgiu, jeigu tai iš viso yra įmanoma. Tačiau mokslininkai vis dar ginčijasi dėl kibernetinio karo ir kibernetines erdvės terminologijos, tuo tarpu kibernetinio karo grėsme tik didėja. Pasaulio valstybes tai suprasdamos bando ištaisyti teisės trukumus, tačiau nesėkmingai. Bet kokie pasiūlymai ir susitarimai galioja tik nedideliam valstybių ratui. O didžiosios valstybės tuo tarpu nenoriai atsisakytu savo kibernetinio pajėgumo. Darbe buvo išanalizuotos tarptautinės sutartys bei valstybių praktika bandant pritaikyti esamus režimus kibernetinio karo reguliacijai. Darytinos išvados, kad esamos tarptautinės teisinės bazės taikymas geriausiu atveju yra sudėtingas ir nenatūralus. Realybė yra tai, kad kibernetiniam karui netinka nei vienas režimas. O toks jo taikymas, deja sukelia daugiau problemų nei buvo prieš tai. Tačiau dar nėra išsemtos visos galimybės ir ateitis gali parodyti teisingą sprendimą. Tuo tarpu valstybes yra pasiruošusios vesti derybas dėl tarptautinės... [toliau žr. visą tekstą]

Nesta tese investigam-se três pontos ligados a fragilidades topológicas de grafos e suas aplicações a redes complexas reais e, em especial, a redes de relacionamentos criminais. Na primeira etapa, apresenta-se in abstracto um método inédito e eficiente de fragmentação de redes complexas por módulos. O procedimento identifica em primeiro lugar comunidades topológicas por meio da qual a rede pode ser representada usando algoritmos heurísticos de extração de comunidades. Então, somente os nós que participam de ligaçõees inter-comunitaárias são removidos em ordem decrescente de sua centralidade de intermediação. Ilustra-se o método pela aplicação a uma variedade de redes reais nas áreas social, de infraestrutura, e biológica. Mostra-se que a abordagem por módulos supera ataques direcionados a vértices baseados somente no ordenamento de índices de centralidade, com ganhos de eficiência fortemente relacionados à modularidade da rede.No segundo momento, introduzem-se os conceitos de robustez e fragilidade de redes generalizadas para avaliar o quanto um determinado sistema se comporta frente a ataques incompletos. Ainda, avalia-se o desempenho (relação entre robustez e custo computacional) de diversos ataques sequenciais e simultâneos a redes modulares por meio de uma medida empírica que chamamos de performance. Mostra-se por meio de redes artificiais de referência e de redes reais que para sistemas altamente modulares a estratégia de fragmentação por módulos apresenta um desempenho até 10 vezes superior aos demais ataques. Na última etapa, explora-se com maior profundidade a natureza subjacente de redes reais de relacionamentos criminais. Apresenta-se uma rede única e sem precedentes construída pela Polícia Federal Brasileira consistindo de mais de 35.000 relacionamentos entre 24.000 indivíduos. Os dados foram coletados entre abril e agosto de 2013 e consistem em informações fornecidas diretamente pelos investigadores responsáveis de cada caso. O sistema apresenta características típicas de redes sociais, porém é bem mais “escuro"que o comportamento típico, com baixos níveis tanto de densidade de arestas quanto de eficiência de rede. Além do mais, o sistema é extremamente modular o que implica ser possível desmantelar toda a rede de crimes federais brasileiros com a remoção de aproximadamente 2% dos indivíduos escolhidos conforme a prescrição do método modular. Também, a rede é controlável no sentido da teoria matemática de controle, significando que com acesso a aproximadamente 20% dos nós é possível, em tese, levar qualquer variável dinâmica de um estado inicial a um estado final arbitrário em um tempo finito. Exibi-se tambám uma análise topológica e de fragilidades de uma segunda rede criminal relacionada a investigações da Polícia Federal. Trata-se de um fórum online destinado à prática de crimes cibernéticos na chamada camada profunda da internet (deep web). (Continuação ) Após a coleta dos dados foi possível construir uma rede de relacionamentos com quase 10.000 indivíduos. Comparou-se, entãoo, a estratégia usada de fato pela Polícia Federal durante a Operação Darknet com a previsão teórica de ataques topológicos à rede criminal e mostrou-se que ataques dirigidos por grau teriam fragmentado o sistema de maneira quase 15 vezes mais eficiente. Por outro lado, esta rede não é modular apesar de novamente apresentar uma arquitetura mais “escura" que o usual. Por termo, demonstra-se que os ataques por arestas estão diretamente relacionados ao aprisionamento enquanto que a ressocialização e/ou morte dos indivíduos é melhor interpretada como a remoção por vértices. Destarte, comprovou-se que de um ponto de vista topológico a ressocialização é de fato mais eficiente em reduzir a criminalidade do que o aprisionamento. Contudo, na rede de crimes federais estudada essa diferenca é muito pequena, de tal modo que ambas as políticas poderiam, em tese, ser aplicadas a fim de se combater eficientemente o sistema criminoso. / In this thesis we investigate three points connected to topological fragilities of graphs and their applications to real complex networks and, in particular, to networks of criminal relationships. In the first step, we present an unprecedented and efficient method of fragmentation of complex networks by modules. Firstly, the procedure identifies topological communities through which the network can be represented using heuristic communities extraction algorithms. After that, only the nodes that bridge communities are removed in descending order of their betweenness centrality . We illustrate the method by the applying it to a variety of real networks in the social, infrastructure, and biological fields. We show that the modular approach outperforms attacks traditional attacks based only on the ordering of centrality indexes, with efficiency gains strongly related to the modularity of the network. In the second moment, we introduce the concepts of generalized robustness and fragility of networks to evaluate how much a certain system behaves in the face of incomplete attacks. Also, we evaluate the relation between robustness and computational cost of several sequential and simultaneous attacks to modular networks by means of an empirical measure that we call performance. In this sense, we show through artificial and real networks that for highly modular systems the strategy of fragmentation by modules presents a performance up to 10 times superior to traditional attacks. In the last step, we explore in more depth the underlying nature of real networks of criminal relationships. We present a unique and unprecedented network built by the Brazilian Federal Police consisting of more than 35,000 relationships among 24,000 individuals. The data were collected between April and August 2013 and consist of information provided directly by the investigators responsible for each case. The system has typical characteristics of social networks, but is much "darker"than traditional social networks, with low levels of edge density and network efficiency. Moreover, the network is extremely modular which implies that it is possible to dismantle all the network of Brazilian federal crimes with the removal of approximately 2% of the individuals chosen according to the modular method. Also the network is controllable in the sense of the mathematical control theory, meaning that with access only to 20% of nodes it is possible, In theory, to take any dynamic variable from an initial state to an arbitrary final state in a finite time. We also show a topological analysis of a second criminal network related to Federal Police investigations. This is an online forum for cybercrime in the so-called deep web. After the data collection, it was possible to build a network of relationships with almost 10,000 individuals. We then compared the strategy actually used by the Federal Police during Operation Darknet with the theoretical prediction of topological attacks on the criminal network and showed that degree-based attacks would have fragmented the system almost 15 times more efficiently. On the other hand, this network is not modular despite presenting a "darker"architecture than usual. As a last result, this particular system is not controllable in practical terms. We finish the study by showing that edge attacks are directly related to the imprisonment whereas the resocialization and/or death of the individuals is better interpreted as the removal of vertices. Thus, we prove that from a topological point of view resocialization is in fact more efficient in reducing crime rates than imprisonment. However, in the network of federal crimes studied here this difference is very small, so that both policies could in theory be applied in order to combat effectively the criminal system.

Nesta tese investigam-se três pontos ligados a fragilidades topológicas de grafos e suas aplicações a redes complexas reais e, em especial, a redes de relacionamentos criminais. Na primeira etapa, apresenta-se in abstracto um método inédito e eficiente de fragmentação de redes complexas por módulos. O procedimento identifica em primeiro lugar comunidades topológicas por meio da qual a rede pode ser representada usando algoritmos heurísticos de extração de comunidades. Então, somente os nós que participam de ligaçõees inter-comunitaárias são removidos em ordem decrescente de sua centralidade de intermediação. Ilustra-se o método pela aplicação a uma variedade de redes reais nas áreas social, de infraestrutura, e biológica. Mostra-se que a abordagem por módulos supera ataques direcionados a vértices baseados somente no ordenamento de índices de centralidade, com ganhos de eficiência fortemente relacionados à modularidade da rede.No segundo momento, introduzem-se os conceitos de robustez e fragilidade de redes generalizadas para avaliar o quanto um determinado sistema se comporta frente a ataques incompletos. Ainda, avalia-se o desempenho (relação entre robustez e custo computacional) de diversos ataques sequenciais e simultâneos a redes modulares por meio de uma medida empírica que chamamos de performance. Mostra-se por meio de redes artificiais de referência e de redes reais que para sistemas altamente modulares a estratégia de fragmentação por módulos apresenta um desempenho até 10 vezes superior aos demais ataques. Na última etapa, explora-se com maior profundidade a natureza subjacente de redes reais de relacionamentos criminais. Apresenta-se uma rede única e sem precedentes construída pela Polícia Federal Brasileira consistindo de mais de 35.000 relacionamentos entre 24.000 indivíduos. Os dados foram coletados entre abril e agosto de 2013 e consistem em informações fornecidas diretamente pelos investigadores responsáveis de cada caso. O sistema apresenta características típicas de redes sociais, porém é bem mais “escuro"que o comportamento típico, com baixos níveis tanto de densidade de arestas quanto de eficiência de rede. Além do mais, o sistema é extremamente modular o que implica ser possível desmantelar toda a rede de crimes federais brasileiros com a remoção de aproximadamente 2% dos indivíduos escolhidos conforme a prescrição do método modular. Também, a rede é controlável no sentido da teoria matemática de controle, significando que com acesso a aproximadamente 20% dos nós é possível, em tese, levar qualquer variável dinâmica de um estado inicial a um estado final arbitrário em um tempo finito. Exibi-se tambám uma análise topológica e de fragilidades de uma segunda rede criminal relacionada a investigações da Polícia Federal. Trata-se de um fórum online destinado à prática de crimes cibernéticos na chamada camada profunda da internet (deep web). (Continuação ) Após a coleta dos dados foi possível construir uma rede de relacionamentos com quase 10.000 indivíduos. Comparou-se, entãoo, a estratégia usada de fato pela Polícia Federal durante a Operação Darknet com a previsão teórica de ataques topológicos à rede criminal e mostrou-se que ataques dirigidos por grau teriam fragmentado o sistema de maneira quase 15 vezes mais eficiente. Por outro lado, esta rede não é modular apesar de novamente apresentar uma arquitetura mais “escura" que o usual. Por termo, demonstra-se que os ataques por arestas estão diretamente relacionados ao aprisionamento enquanto que a ressocialização e/ou morte dos indivíduos é melhor interpretada como a remoção por vértices. Destarte, comprovou-se que de um ponto de vista topológico a ressocialização é de fato mais eficiente em reduzir a criminalidade do que o aprisionamento. Contudo, na rede de crimes federais estudada essa diferenca é muito pequena, de tal modo que ambas as políticas poderiam, em tese, ser aplicadas a fim de se combater eficientemente o sistema criminoso. / In this thesis we investigate three points connected to topological fragilities of graphs and their applications to real complex networks and, in particular, to networks of criminal relationships. In the first step, we present an unprecedented and efficient method of fragmentation of complex networks by modules. Firstly, the procedure identifies topological communities through which the network can be represented using heuristic communities extraction algorithms. After that, only the nodes that bridge communities are removed in descending order of their betweenness centrality . We illustrate the method by the applying it to a variety of real networks in the social, infrastructure, and biological fields. We show that the modular approach outperforms attacks traditional attacks based only on the ordering of centrality indexes, with efficiency gains strongly related to the modularity of the network. In the second moment, we introduce the concepts of generalized robustness and fragility of networks to evaluate how much a certain system behaves in the face of incomplete attacks. Also, we evaluate the relation between robustness and computational cost of several sequential and simultaneous attacks to modular networks by means of an empirical measure that we call performance. In this sense, we show through artificial and real networks that for highly modular systems the strategy of fragmentation by modules presents a performance up to 10 times superior to traditional attacks. In the last step, we explore in more depth the underlying nature of real networks of criminal relationships. We present a unique and unprecedented network built by the Brazilian Federal Police consisting of more than 35,000 relationships among 24,000 individuals. The data were collected between April and August 2013 and consist of information provided directly by the investigators responsible for each case. The system has typical characteristics of social networks, but is much "darker"than traditional social networks, with low levels of edge density and network efficiency. Moreover, the network is extremely modular which implies that it is possible to dismantle all the network of Brazilian federal crimes with the removal of approximately 2% of the individuals chosen according to the modular method. Also the network is controllable in the sense of the mathematical control theory, meaning that with access only to 20% of nodes it is possible, In theory, to take any dynamic variable from an initial state to an arbitrary final state in a finite time. We also show a topological analysis of a second criminal network related to Federal Police investigations. This is an online forum for cybercrime in the so-called deep web. After the data collection, it was possible to build a network of relationships with almost 10,000 individuals. We then compared the strategy actually used by the Federal Police during Operation Darknet with the theoretical prediction of topological attacks on the criminal network and showed that degree-based attacks would have fragmented the system almost 15 times more efficiently. On the other hand, this network is not modular despite presenting a "darker"architecture than usual. As a last result, this particular system is not controllable in practical terms. We finish the study by showing that edge attacks are directly related to the imprisonment whereas the resocialization and/or death of the individuals is better interpreted as the removal of vertices. Thus, we prove that from a topological point of view resocialization is in fact more efficient in reducing crime rates than imprisonment. However, in the network of federal crimes studied here this difference is very small, so that both policies could in theory be applied in order to combat effectively the criminal system.