Título: Smart Network Slicing - Otimização de Recursos em Ambientes 5G
Autor: Giselle Goiz da Silva
Resumo: O advento das redes 5G vai permitir que as cidades estejam cada vez mais conectadas e inteligentes, e têm o potencial de alavancar o paradigma de Internet das Coisas (IoT - Internet of Things), que será uma realidade cada vez mais presente no dia a dia da sociedade. Prover recursos que atendam as mais diversas finalidades de serviços, com requisitos de negócios e SLAs bastante específicos, tais como: Conexões massivas e cirurgias remotas, necessitam de dimensionamento adequado dos recursos.
Tecnologias como o Network Slicing que permite a segmentação lógica da rede, a fim de garantir QoS e QoE, End-to-end para os diferentes modelos de negócio, são fundamentais.
Ocorre que, ao manter recursos dedicados a cada slice eventualmente resultará em subutilização dos mesmos, pois a demanda é variável. Diante disso, o compartilhamento eficiente da infraestrutura de rede torna-se um grande desafio para as operadoras de telecomunicações.
O desenvolvimento de tecnologias para automação de sistemas, aliados a técnicas de Machine Learning, podem tornar o processo de adequação da rede mais rápido e flexível.
O modelo de solução deste trabalho intitulado Smart Network Slicing (SNS), utiliza algoritmos de Machine Learning (ML) para classificação das aplicações e previsão do comportamento do tráfego, com base nos fluxos de rede. O sistema inteligente visa automatizar o processo de alocação de recursos, adequando proativamente as configurações dos slices de acordo com a necessidade prevista pelo algoritmo. Este trabalho tem por objetivo otimizar a alocação de recursos entre network slices e proporcionar um compartilhamento de infraestrutura eficiente, evitando congestionamentos e resultando em um aumento da disponibilidade dos serviços para as redes 5G.
Título: SPED - SFC Placement in Edge-Cloud Environment: a Distributed Approach
Autor: Anselmo Luiz Éden Battisti
Resumo: In the last decades, there has been a trend to virtualize computing and networking resources over the physical infrastructure. This virtualization paradigm first reaches the core functions of the networks like NATs, Firewalls, and Deep Packet Inspectors, thus creating the NFV paradigm. Nowadays, even high-level functions directly consumed by the users like CDN, video encoders, and anti-virus are also virtualized. The adoption of the NFV paradigm reduces the space needed for network hardware, decreases network power consumption, diminishes network maintenance costs, and increases the life cycles of network hardware, facilitating the management and orchestration of the network functions. With the adoption of the NFV paradigm, a new problem named VNF Placement emerged. The VNF Placement defines the computational node that will be used to execute the VNF. However, virtualizing only VNFs individually is insufficient to meet the user's services requirements even in network traffic demands as service chaining complexity. These demands have created a new paradigm called SFC (Service Function Chain). The SFC is a chain of multiple VNFs with a set of SLAs. Now the SFC Placement Problem has emerged, and new challenges must be addressed beyond defining where to execute each VNF of the requested SFC, the networking connectivity must be created to enable the flow across all the VNFs. With the advance of the Edge-Cloud computing paradigm, the VNFs of the SFC can be executed in multiple nodes that can potentially be managed by different service providers. In this scenario, a centralized placement for the SFC is inadequate, given the huge number of nodes in the environment and the confidentiality problems. Thus, a distributed mechanism should be created to allocate the requested SFC without complete knowledge of the above infrastructure where the VNFs will be executed. In this work, we propose a new approach, named SPED, to solve the SFC Placement problem in a multi-domain environment in a distributed fashion. The solution encompasses the architectural components, a new mapping model between the SFC Placement Problem and the Game-Theory approach, besides a new data aggregation pattern over the network topology.
Título: L-PRISM - Language specification for the creation of multimedia service chains based on virtualization of multimedia sensors
Autor: Franklin Jordan Ventura Quico
Resumo: Multimedia service chains are combinations of two or more components that process one or more multimedia streams (audio, video, image) to perform some actions on this multimedia stream and finally distribute its result. Virtualization technology makes the development processes of different applications less costly to implement and, in turn, makes the administration of its components more accessible. Network Function Virtualization (NFV) is a concept that has become very important in recent years since different applications, architectures, language specifications, etc., have been developed based on this concept. That makes the development or use of applications based on virtualization more accessible. Over time, the community has developed different multimedia applications using virtualization concepts. Most of these applications are created without specifications for their use or configuration. The lack of a specification in the applications means that external users need knowledge of programming languages, software architectures, or different technologies to understand and use these applications. This work focuses on developing a domain language specification that helps more manageable and more efficiently develop multimedia service chains based on Virtual Network Functions (VNFs) of multimedia type. The application of virtualization technology in multimedia applications gives rise to a new concept called Virtualized Multimedia Sensors (VMS). L-PRISM is a domain-specific language based on YAML. L-PRISM offers developers to register their virtualized multimedia applications to orchestrate multimedia service chains in distributed environments based on VMS.
