A comunicação Dispositivo-a-Dispositivo (D2D) é um dos mecanismos para a desoneração das redes 4G, via offloading de dados para outras redes. Redes híbridas D2D/4G possuem como desafio a multiplexação eficiente das interfaces de comunicação, o que pode impactar diretamente na Qualidade de Experiência (QoE) dos usuários. Este trabalho apresenta o QD4G, uma plataforma que emprega Aprendizado de Máquina para controlar quando clientes usando o padrão DASH (Dynamic Adaptative Streaming over HTTP) devem empregar a rede 4G ou uma rede D2D. O objetivo é melhorar o QoE ao proporcionar a transmissão de vídeo em uma maior resolução. Testes foram realizados com dispositivos reais e os resultados mostraram que a predição permite atingir até 87% de acerto, com um aumento na resolução média do vídeo entre os usuários via D2D de até 150%, ao mesmo tempo desonerando o tráfego de dados da rede 4G em até 80% nos cenários avaliados.
One key component for efficient opportunistic device-to-device (D2D) deployment is cache management. It determines which content to store opportunistic D2D communications. Existing solutions focus on the nature of content or mobility attributes, but most of them neglect their joint influence. Moreover, most solutions rely on a preloading phase, filling caches with content that the respective users may not consume, but that may be of interest to other nodes, and increasing traffic overhead in the core network. Further, a popular file may be a lousy candidate for opportunistic D2D because contact opportunities may not provide enough transfer capacity. To solve this issue, we propose a model that computes priority values based on both content and mobility attributes. Our approach considers only files that users have consumed, therefore eliminating a preloading phase. Using real-world and synthetic mobility traces, we compare our solution with Least Recently Stored replacement, as well as a state-of-the-art approach that also considers content and mobility attributes. Results show an increase in the global cache hit rate of almost 80% in scenarios that offer many files, and of around 420% in scenarios with a few users. The priority model generates 90% lower overhead in terms of the control bytes. We also apply our solution in a chunk-based adaptive video streaming application. We observe that our solution leads to higher video delivery ratios when compared to the baselines.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.