Comunicação dispositivo-a-dispositivo orientada a atributos para descarregamento de dados em redes celulares

Abstract

Data traffic from cellular networks shows an exponential growth year after year. Opportunistic mobile data offloading alleviates the problem by moving traffic to other networks, and Device-To-Device (D2D) communications can be used for such a purpose. In this way, users that request the same data can share it among themselves through a direct link, thus eliminating the need of multiple requests in the cellular network. Historically, most proposals of a D2D/cellular infrastructure do not consider the tradeoff between device mobility and the transmitted content. Another drawback is that these proposals rely on content preloading. This forces the device to cache content that might not be of interest of the respective user, but only of his neighbors, increasing the traffic on the network. To increase the use of opportunistic D2D communications and the device cache, this way reducing the traffic on the cellular network, this work proposes two priority models for cache replacement, computing priority values for the stored content. The cache replacement procedure is based on monitored data related to the mobility of each device, as well as their respective content download profile. Such a procedure is distributed among the devices, occurring every time they receive new content, and their respective cache is overloaded. The proposed models consider only content consumed and stored by each device, eliminating content preloading. The first model, called General Purpose Priority Model (MP-PG), considers scenarios with generic content of fixed quality that cannot be fragmented. The second model, called Quality-Aware Priority Model (MP-CQ), considers content splittable into segments (or chunks) of variable quality (e.g., adaptive video). MP-CQ also improves the content quality perceived by each user. The performance evaluation of MP-PG and MP-CQ occurred through simulations, varying parameters like the amount of content and users in the network. Both models were compared with a classic solution that removes the oldest content, hence not considering the joint relationship between mobility and content attributes. The proposed models were also compared with a state-of-the-art solution. Results prove the effectiveness of both models in terms of the use of D2D communications and the cache, as well as the cost related to control messages.