A research report submitted in partial fulfilment to the Degree of Master of Commerce (Information Systems) in the school of Economic and Business Sciences, University of the Witwatersrand, 2018 / Purpose: The aim in this research study was to explore the role and competencies of Data Scientists in South Africa as the role starts to emerge. Due to the newness of the role, jobs in this sphere are currently being filled by skilled professionals moving from other related areas. Knowledge and skills for Data Scientists were explored in order to examine the role of a Data Scientist and the competencies they should have. Design/methodology/approach: The studies that have been published on the role of a Data Scientist are limited as the field of Data Science is still new. Therefore the design of the research was exploratory and used qualitative methods. Data gathered for this research was analysed using thematic analysis. The study used respondents drawn from the banking and insurance industries as they are amongst the first to employ Data Scientists in the real sense of the term in South Africa. Six Data Scientists were interviewed. Originality/value: Research that focuses on the role of Data Scientists especially in South Africa is limited as most of the research has taken place in developed countries. There is also limited research on the role of a Data Scientist within the banking and insurance industry. This study contributes to practitioner and research knowledge by exploring the emerging role of a Data Scientist in the South African context. Practical implications: This research improves our understanding of the knowledge and skills Data Scientists should have within the banking and insurance industry. This research adds insight by highlighting the role that Data Scientists are currently undertaking by providing information on the specific skills that they report as required. This research can help in the shaping of education and developing the required skills for individuals who intend to pursue the career path of a Data Scientist as well as help managers hire the right people for the position of a Data Scientist. / TL2019

No description available.

Les travaux de cette thèse s'inscrivent dans le contexte général des réseaux logiciels, dits "Software-Defined Networking" (SDN). Ce paradigme récent est l'une des initiatives les plus notables pour rendre les réseaux actuels programmables ou, en d'autres termes, plus simple à configurer, à tester, à corriger et à faire évoluer. Dans un écosystème SDN, l'interface nord (Northbound API) est utilisée par l'administrateur réseaux pour définir ses politiques et programmer le plan de contrôle, elle représente donc un enjeu majeur. Idéalement, cette interface nord devrait permettre aux administrateurs de décrire, le plus simplement possible, des services réseaux et leurs interactions, plutôt que de spécifier comment et sur quels équipements physiques ils doivent être déployés. Des travaux existants montrent que cela peut être notamment réalisé grâce à des solutions de virtualisation de réseaux et des langages de programmation dédiés de haut niveau. L'objectif de ce travail de thèse est de proposer une nouvelle interface nord qui, d'une part, exploiterait la virtualisation de réseau et, d'autre part, exposerait ses services sous la forme d'un langage de programmation dédié. Actuellement, plusieurs langages intégrant des solutions de virtualisation de réseau existent. Néanmoins, nous pensons que les modèles d'abstraction qu'ils utilisent pour construire des réseaux virtuels restent inappropriés pour assurer des critères de simplicité, modularité et flexibilité des topologies virtuelles et des programmes de contrôle. Dans ce contexte, nous proposons un nouveau langage de contrôle de réseaux nommé AirNet. Ce dernier intègre un modèle d'abstraction dont la principale caractéristique est d'offrir une séparation nette entre les équipements de bordure (Edge) et de cœur de réseau (Fabric). Cette idée est bien connue et acceptée dans le domaine des architectures réseaux. L'originalité de notre contribution étant de faire remonter ce concept au niveau du plan de contrôle virtuel et non de le restreindre au seul plan physique. Ainsi, des frontières logiques entre les différents types de politiques existeront (fonctions de contrôle et de données vs. fonctions de transport), garantissant ainsi la modularité et la réutilisabilité de tout ou partie du programme de contrôle. De plus, dans l'approche proposée, la définition du réseau virtuel et des politiques peut être totalement dissociée de l'infrastructure physique cible, favorisant ainsi la portabilité des applications de contrôle. Une implémentation du langage AirNet a également été réalisée. Ce prototype inclut en particulier une bibliothèque des primitives et opérateurs du langage, ainsi qu'un hyperviseur qui assure la composition des politiques de contrôle sur un réseau virtuel, et leur transposition (mapping) sur l'infrastructure physique. Afin de s'appuyer sur des contrôleurs SDN existants, l'hyperviseur inclut des modules d'intégration des contrôleurs POX et RYU. Une validation expérimentale a été menée sur différents cas d'étude (filtrage, répartition de charge, authentification dynamique, limitation de bande passante, etc.) dont les résultats attestent de la faisabilité de la solution. Enfin, des mesures de performances ont montré que le surcoût apporté par cette nouvelle couche d'abstraction est parfaitement acceptable. / The work of this thesis falls within the general context of software-defined networking (SDN). This new paradigm is one of the most significant initiatives to enable networks programmability or, in other words, to make current networks easier to configure, test, debug and evolve. Within an SDN ecosystem, the Northbound interface is used by network administrators to define policies and to program the control plane, it thus represents a major challenge. Ideally, this northbound interface should allow administrators to describe, as simply as possible, network services and their interactions, rather than specifying how and on what physical device they need to be deployed. Current related works show that this can be partly achieved through virtualization solutions and high-level domain specific languages (DSL). The objective of this thesis is to propose a new Northbound interface which will, on the one hand, rely on network virtualization and, on the other hand, expose its services as a domain specific programming language. Currently, several languages that include network virtualization solutions exist. Nevertheless, we believe that the abstract models they are using to build virtual networks remain inadequate to ensure simplicity, modularity and flexibility of virtual topologies and control programs. In this context, we propose a new network control language named AirNet. Our language is built on top of an abstraction model whose main feature is to provide a clear separation between edge and core network devices. This concept is a well-known and accepted idea within the network designer community. The originality of our contribution is to lift up this concept at the virtual control plane, not limiting it solely at the physical plane. Thus, logical boundaries between different types of policies will exist (control and data functions vs. transport functions), ensuring modularity and reusability of the control program. Moreover, in the proposed approach, the definition of the virtual network and policies is totally dissociated from the target physical infrastructure, promoting the portability of control applications. An implementation of the AirNet language has also been done. This prototype includes in particular a library that implements the primitives and operators of the language, and a hypervisor that achieves the composition of the control policies on the virtual network, and their mapping on the physical infrastructure. In order to rely on existing SDN controllers, the hypervisor includes integration modules for the POX and RYU controllers. An experimental validation has been also conducted on different use cases (filtering, load balancing, dynamic authentication, bandwidth throttling, etc.), whose results demonstrate the feasibility of our solution. Finally, performance measurements have shown that the additional cost brought by this new abstraction layer is perfectly acceptable.

Denna studie fokuserar på att samla information om Software Defined Networking, dess protokoll och dess kontroller. Det som jag har lärt mig under arbetet kommer att användas för att utvärdera två olika kontroller, POX och ONOS. Ett traditionellt nätverks kommer att sättas upp fysiskt och användas som en grund för att jämföra kontrollerna. Den traditionella lösningen använder två routrar och fyra switchar, och egenskaper som testas är bland annat lager 2 och lager 3 samt deras protokoll för redundans. Kontrollerna kommer sedan att användas för att se om de lever upp till samma krav. Resultaten av denna studie visar att varken POX eller ONOS kunde användas för varje testat scenario, inte med de moduler som kontrollerna kommer förinstallerade med. Det visade också att de egenskaper som de levde upp till var en hel del lättare att konfigurera och övervaka jämfört med dess traditionella motsvarigheter. Detta visar vikten av att lista ut vad som behövs och förväntas från nätverket innan man försöker hitta en passande lösning för att utföra detta. All information som samlats i denna studie används också för att skapa en laboration som ska introducera andra till koncepten kring SDN. Den undersöker hur Mininet kan användas för att virtualisera ett nätverk, hur flöden kan installeras med OpenFlow samt hur en kontroller kan användas för att förenkla administration av ett nätverk. / This study focuses on gathering information about Software Defined Networking, it's protocols ans it's controllers. What I have learned doing this will be used to evaluate two different controllers, POX and ONOS. A traditional network setup will be set up physically and serve as a base when it comes to comparing the controllers. The traditional setup includes two routers and four switches, and among the tested characteristics are layer 2 and 3 and it's redundancy protocols. The controllers will then be used to try and live up to the same characteristics. The result of this study shows that neither POX nor ONOS could be used for every scenario tested, not with the basic modules the controllers comes with. It also showed that the characteristics they did manage was a fair bit easier to setup and monitor compared to it's traditional counterparts, thus showing the importance of figuring out what is needed from a network before trying to find a fitting solution to how it needs to be set up. All the information gathered in this study is also used to create a lab instruction meant to introduce others to the concepts of SDN. It explores how to use Mininet to virtualise a network environment, how to install flows using OpenFlow and how to use a controller to simplify the management of the network.

Layer 4-7 network functions (NF), such as Firewall or NAPT, have traditionally been implemented in specialized hardware with little to no programmability and extensibility. The scientific community has focused on realizing this functionality in software running on commodity servers instead. Despite the many advancements over the years (e.g., network I/O accelerations), software-based NFs are still unable to guarantee some key service-level objectives (e.g., bounded latency) for the customer due to their reactive approach to workload changes. This thesis argues that Machine Learning techniques can be utilized to forecast how traffic patterns change over time. A network orchestrator can then use this information to allocate resources (network, compute, memory) in a timely fashion and more precisely. To this end, we have developed Mantis, a control plane network application which (i) monitors all forwarding devices (e.g., Firewalls) to generate performance-related metrics and (ii) applies predictors (moving average, autoregression, wavelets, etc.) to predict future values for these metrics. Choosing the appropriate forecasting technique for each traffic workload is a challenging task. This is why we developed several different predictors. Moreover, each predictor has several configuration parameters which can all be set by the administrator during runtime. In order to evaluate the predictive capabilities of Mantis, we set up a test-bed, consisting of the state-of-the-art network controller Metron [16], a NAPT NF realized in FastClick [6] and two hosts. While the source host was replaying real-world internet traces (provided by CAIDA [33]), our Mantis application was performing predictions in real time, using a rolling window for training. Visual inspection of the results indicates that all our predictors have good accuracy, excluding (i) the beginning of the trace where models are still being initialized and (ii) instances of abrupt change. Moreover, applying the discrete wavelet transform before we perform predictions can improve the accuracy further. / Nätverksfunktioner i lager 4-7 som t.ex. brandväggar eller NAPT har traditionellt implementeras på specialdesignad hårdvara med väldigt få programeringsegenskaper. Forskning inom datakomunikation har fokuserat på att istället möjliggöra dessa funktioner i mjukvara på standardhårdvara. Trots att många framsteg har gjorts inom området under de senaste åren (t.ex. nätverks I/O accelerering), kan inte mjukvarubaserade nätverksfunktioner garantera önskad tjänstenivå för kunderna (t.ex. begränsade latensvärden) p.g.a. det reaktiva tillvägagångsättet när arbetslasten ändras. Den här avhandlingen visar att med hjälp av maskininlärning så går det att förutse hur trafikflöden ändras över tid. Nätverksorkestrering kan sedan användas för att allokera resurser (bandbredd, beräkning, minne) i förväg samt mer precist. För detta ändamål har vi utvecklat Mantis, en nätverksapplikation i kontrolplanet som övervakar alla nätverksenheter för att generera prestandabaserade mätvärden och använder matematiska prediktorer (moving average, autoregression, wavelets, o.s.v.) för att förutse kommande ändringar i dessa värden. Det är en utmaning att välja rätt metod för att skapa prognosen för varje resurs. Därför har vi utvecklat flera olika prediktorer. Dessutom har varje prediktor flera konfigurationsvärden som kan ändras av administratören. För att utvärdera Mantis prognoser har vi satt upp ett testnätverk med en av marknadens ledande nätverkskontrollers, Metron [16], en NAPT nätverksfunktion implementerad med FastClick [6] och två testnoder. Den ena noden skickar data hämtad från verklig Internettrafik (erhållen från CAIDA [33]) samtidigt som vår applikation, Mantis, skapar prognoser i realtid. Manuell inspektion av resultaten tyder på att alla våra prediktorer har god precision, förutom början av en spårning då modellerna byggs upp eller vid abrupt ändring. Dessutom kan precisionen ökas ytterligare genom att använda diskret wavelet transformering av värdena innan prognosen görs.

Providing performance and isolation guarantees for applications running in virtualized datacenter environments requires continuous management of the underlying physical resources. For communication- and I/O-intensive applications running on such platforms, the management methods must adequately deal with the shared use of the high-performance fabrics these applications require. In particular, new classes of latency-sensitive and data-intensive workloads running in virtualized environments rely on emerging fabrics like 40+Gbps Ethernet and InfiniBand/RoCE with support for RDMA, VMM-bypass and hardware-level virtualization (SR-IOV). However, the benefits provided by these technology advances are offset by several management constraints: (i) the inability of the hypervisor to monitor the VMs’ usage of these fabrics can affect the platform’s ability to provide isolation and performance guarantees, (ii) the hypervisor cannot provide fine-grained I/O provisioning or perform management decisions for VMs, thus reducing the degree of consolidation that can be supported on the platforms, and (iii) without such support it is harder to integrate these fabrics into emerging cloud computing platforms and datacenter fabric management solutions. This is made particularly challenging for workloads spanning multiple VMs, utilizing physical resources distributed across multiple server nodes and the interconnection fabric. This thesis addresses the problem of realizing a flexible, dynamic resource management system for virtualized platforms with high performance fabrics. We make the following key contributions: (i) A lightweight monitoring tool, IBMon, integrated with the hypervisor to monitor VMs’ use of RDMA-enabled virtualized interconnects, using memory introspection techniques. (ii) The design and construction of a resource management system that leverages IBMon to provide latency-sensitive applications performance guarantees. This system is built on microeconomic principles of supply and demand and can be deployed on a per-node (Resource Exchange) or a multi-node (Distributed Resource Exchange) basis. Fine-grained resource allocations can be enforced through several mechanisms, including CPU capping or fabric-level congestion control. (iii) Sphinx, a fabric management solution that leverages Resource Exchange to orchestrate network and provide latency proportionality for consolidated workloads, based on user/application-specified policies. (iv) Implementation and experimental evaluation using InfiniBand clusters virtualized with the Xen or KVM hypervisor, managed via the OpenFloodlight SDN controller, and using representative data-intensive and latency-sensitive benchmarks.

Em um futuro próximo, “dispositivos inteligentes” serão massivamente empregados em campos de batalha. Essa já é uma realidade, porém, o número de dispositivos utilizados em campos de batalha tende a aumentar em ordens de magnitude. As redes de comunicação de dados serão essenciais para transmitir os dados que esses dispositivos coletam e transformá-los em informações valiosas utilizadas como suporte à atuação humana. O suporte à tomada de decisão, ou mesmo níveis de autonomia, permitindo que estes dispositivos coordenem outros dispositivos, exigem comunicação contínua. Desafios relacionados à comunicação surgirão devido à dinamicidade do ambiente. A configuração da rede deve refletir decisões superiores automaticamente. A grande escala das redes conectando os altos escalões, tropas, veículos e sensores, aliada à falta de padronização dos dispositivos, tornará a integração destes desafiadora. Em um ambiente tão heterogêneo, muitos protocolos e tecnologias coexistirão. As redes de campo de batalha são um elemento de suma importância nas operações militares modernas e conceito de guerra centrada em rede é uma tendência sem volta e influencia desde os altos escalões até o controle de tropas Embora estudos tenham sido realizados nessa área, a maioria deles aborda redes estratégicas de alto nível e portanto não levam em conta as “redes táticas de última milha” (TEN), que compreendem dispositivos de comunicação com recursos limitados, como sensores ou ainda pequenos veículos aéreos não tripulados. Em uma tentativa de preencher esta lacuna, esse trabalho propõe uma arquitetura que combina conceitos dos paradigmas de redes definidas por software (SDN) juntamente com redes tolerantes à atraso/disrupçoes (DTN), para aplicação em redes táticas de última milha. O uso de SDN em cenários com nodos móveis é avaliado considerando uma aplicação de vigilância que utiliza streaming de vídeo e medidas de Qualidade de Experiência (QoE) de usuário são coletadas. Com base nos resultados obtidos, uma aplicação em conjunto dos conceitos de SDN e DTN é proposta, além disso abordamos a escolha do nodo que atuará como controlador SDN na rede. Os experimentos foram executados utilizando um emulador de redes. Apesar de pesquisas adicionais serem necessárias – considerado requisitos de segurança, por exemplo – os resultados foram promissores e demonstram a aplicabilidade destes conceitos no cenários das TENs. / The future battlefield tends to be populated by a plethora of “intelligent things”. In some ways, this is already a reality, but in future battlefields, the number of deployed things should be orders of magnitude higher. Networked communication is essential to take real advantage of the deployed devices on the battlefield, and to transform the data collected by them into information valuable for the human warfighters. Support for human decision making and even a level of autonomy, allowing devices to coordinate and interact with each other to execute their activities in a collaborative way require continuous communication. Challenges regarding communication will arise from the high dynamics of the environment. The network adaption and management should occur autonomously, and it should reflect upper-level decisions. The large scale of the network connecting high-level echelons, troops on the field, and sensors of many types, beside the lack of communication standards turn the integration of the devices more challenging. In such a heterogeneous environment, many protocols and communication technologies coexist. This way, battlefield networks is an element of paramount importance in modern military operations Additionally, a change of paradigm regarding levels of autonomy and cooperation between humans and machines is in course and the concept of network-centric warfare is a no way back trend. Although new studies have been carried out in this area, most of these concern higher-level strategic networks, with abundant resources. Thus, these studies fail to take into account the “last-mile Tactical Edge Network (TEN) level,” which comprises resource constrained communication devices carried by troopers, sensor nodes deployed on the field or small unmanned aerial vehicles. In an attempt to fill this gap, this work proposes an architecture combining concepts from software-defined networking (SDN) paradigm and the delay-tolerant approach to support applications in the last-mile TEN. First, the use of SDN in dynamic scenarios regarding node positioning is evaluated through a surveillance application using video streaming and Quality of Experience (QoE) measures are captured on the video player. We also explore the election of nodes to act as SDN Controllers in the TEN environment. The experiments use emulator for SDN with support to wireless networks. Further investigation is required, for example, considering security requirements, however the results are promising and demonstrate the applicability of this architecture in the TEN network scenario.

This research is related to computer networks. In this thesis, three main issues are addressed which affect the performance of any computer network: congestion, efficient resources utilization and link failure. Those issues are related to each other in many situations. Many approaches have been suggested to deal with those issues as well as many solutions were applied. Despite all the improvements of the technology and the proposed solutions, those issues continue to be a burden on the system’s performance. This effect is related to the increase of the Quality of Service (QoS) requirements in modern networks. The basic idea of this research is evolving the intelligence of a conventional computer network when dealing with those issues by adding some features of the Software Defined Networking (SDN). This adoption upgrades the conventional computer network system to be more dynamic and higher self-organizing when dealing with those issues. This idea is applied on a system represented by a computer network that uses the Open Shortest Path First (OSPF) routing protocol. The first improvement deals with the distribution of Internet Protocol (IP) routed flows. The second improvement deals with tunnel establishment that serves Multi-Protocol Label Switching (MPLS) routed flows and the third improvement deals with bandwidth reservation when applying network restoration represented by Fast Re-route (FRR) mechanism to sooth the effect of link failure in OSPF/MPLS routed network. This idea is also applied on another system that uses the Enhanced Interior Gateway Routing Protocol (EIGRP) to improve the performance of its routing algorithm. Adopting the SDN notion is achieved by adding an intelligent controller to the system and creating a dialog of messages between the controller and the conventional routers. This requires upgrading the routers to respond to the new modified system. Our proposed approaches are presented with simulations of different configurations which produce fine results.

Os Pontos de Troca de Tráfego participam de maneira substancial e crítica no ecossistema da Internet, possibilitando conexões entre múltiplos Sistemas Autônomos (ASes, do inglês Autonomous Systems). O gerenciamento das redes de PTT possui como objetivos primários, o gerenciamento dos chamados fluxos elefante (do inglês, elephant flows). Fluxos elefante tendem a existir em número reduzido, porém correspondem à maioria do tráfego em uma infraestrutura de rede. O gerenciamento dos fluxos elefante envolve uma adequada identificação e quando necessário, um redirecionamento destes fluxos para caminhos mais apropriados, de forma a minimizar os possíveis impactos sobre os outros fluxos ativos na rede. Além disso, o gerenciamento de fluxos elefante tornou-se um importante objeto de discussão em PTTs baseados em redes SDN, principalmente porque estas redes dispõem de controladores que possuem uma visão consistente da rede subjacente, o que permite uma gerência destes fluxos de forma refinada. Nesta dissertação, será proposto, desenvolvido e avaliado um sistema de identificação dos fluxos elefante e seus respectivos caminhos de rede, em conjunto com um sistema de recomendação, que possui o objetivo de sugerir configurações alternativas para os fluxos elefante identificados anteriormente nas redes de PTTs baseadas em SDN. Neste sistema, o operador do PTT pode definir templates que em última instância definem como os caminhos dos fluxos elefante serão modificados para atender objetivos específicos. Por fim, será demonstrado que o sistema proposto pode auxiliar o operador do PTT a identificar, gerenciar e mitigar o impacto dos fluxos elefante da rede do PTT. / Internet Exchange Points (IXPs) play a key role in the current Internet architecture enabling cost-effective connections among multiple autonomous systems (ASes). Management of IXP networks is primarily concerned with the management of the so-called elephant flows. Such flows represent a small portion of the total flows of a IXP network but usually have high impact on the overall traffic. Managing elephant flows involves adequate identification and eventually rerouting of such flows to more appropriate locations to minimize the possible negative impact on the other (mice) flows active in the network. Elephant flow management becomes more important in SDN-based IXPs that require controllers to have a consistent view of the underlying network to allow fine-grained adjustment. In this master thesis, we propose, develop, and evaluate an identification system to identify elephant flows and their respectively paths, as well as a recommendation system to suggest alternative configurations to previously identified elephant flows in an SDN-based IXP network. In this solution, the IXP operator can define templates that ultimately define how elephant flows can be reconfigured to achieve a specific objective. We demonstrate that our system can help IXP operators to identify, handle and mitigate the impact of elephant flows in the IXP network.

O projeto de mecanismos flexíveis e eficientes para o posicionamento e encadeamento de funções virtualizadas de rede (VNFs) é essencial para o sucesso de Virtualização de Funções de Rede (Network Function Virtualization, NFV). A maioria das soluções existentes, no entanto, considera custos fixos (e imutáveis) de processamento de fluxos e de largura de banda ao posicionar as VNFs em Pontos de Presença da Rede (N-PoPs). Essa limitação torna-se crítica em redes NFV com fluxos cujos comportamentos são altamente dinâmicos e nas quais os requisitos de processamento e os recursos disponíveis nos NPoPs mudam constantemente. Para preencher essa lacuna, propõe-se o NFV-PEAR, uma plataforma para o posicionamento e encadeamento adaptativo de VNFs. O NFV-PEAR visa (re)organizar periodicamente os posicionamentos e encadeamentos de VNFs previamente determinados, objetivando-se manter um desempenho fim-a-fim aceitável mesmo durante flutuações nos custos de processamento e nos requisitos dos fluxos. Paralelamente, busca-se minimizar as mudanças na rede (por exemplo, a realocação de VNFs ou de fluxos) realizadas para cumprir esse objetivo. Os resultados obtidos, a partir de uma avaliação experimental, mostram que o NFV-PEAR tem potencial para reduzir significativamente o número de mudanças na rede necessárias para assegurar o desempenho fim-a-fim esperado para os fluxos, garantindo assim o funcionamento estável dos serviços. / The design of flexible and efficient mechanisms for proper placement and chaining of virtual network functions (VNFs) is key for the success of Network Function Virtualization (NFV). Most state-of-the-art solutions, however, consider fixed (and immutable) flow processing and bandwidth requirements when placing VNFs in the Network Points of Presence (N-PoPs). This limitation becomes critical in NFV-enabled networks having highly dynamic flow behavior, and in which flow processing requirements and available N-PoP resources change constantly. To bridge this gap, we present NFV-PEAR, a platform for adaptive VNF placement and chaining. In NFV-PEAR, network operators may periodically (re)arrange previously determined placement and chaining of VNFs, with the goal of maintaining acceptable end-to-end flow performance despite fluctuations of flow processing costs and requirements. In parallel, NFV-PEAR seeks to minimize network changes (e.g., reallocation of VNFs or network flows). The results obtained from an experimental evaluation provide evidence that NFV-PEAR has potential to deliver more stable operation of network services, while significantly reducing the number of network changes required to ensure end-to-end flow performance.