DeepFloat: Resource-Efficient Dynamic Management of Vehicular Floating Content

Abstract

Opportunistic communications are expected to play a crucial role in vehicular services that are based on location and require extremely low latency. A widely investigated opportunistic communication paradigm for the local dissemination of contextualized information is Floating Content (FC), which tries to make content float over a geographical area by replicating it whenever two users meet. The key Quality of Service (QoS)indicator for FC is content availability, defined as the fraction of users that received the information that is supposed to float.

Optimizing the use of FC resources while meeting the availability target QoS is a highly complex issue. Fully distributed, distance-based approaches proved to be highly inefficient, and may not meet the target QoS. Centralized, model-based approaches do not perform well in realistic inhomogeneous settings.

In this work, we present a data-driven centralized approach to resource-efficient, QoS-aware dynamic management of FC. We propose a Deep Learning strategy for FC operation, which employs a Convolutional Neural Network (CNN) to capture the relations between the patterns of users mobility, the patterns of content diffusion and replication, and the performance of FC in terms of resource efficiency and availability within a given Zone of Interest (ZOI).

Numerical evaluations show the effectiveness of our approach, as well as the capability of our approach to adapt to mobility pattern changes over time.