Streaming content from a vehicular Cloud

Network densication via small cells is considered as a key step to cope with the data tsunami. Caching data at small cells or even user devices is also considered as a promising way to alleviate the backhaul congestion this densication might cause. However, the former suers from high deployment and maintenance costs, and the latter from limited resources and privacy issues with user devices. We argue that an architecture with (public or private) vehicles acting as mobile caches and communication relays might be a promising middle ground. In this paper, we assume such a vehicular cloud is in place to provide video streaming to users, and that the operator can decide which content to store in the vehicle caches. Users can then greedily ll their playout buer with video pieces of the streamed content from encountered vehicles, and turn to the infrastructure immediately when the playout buer is empty, to ensure uninterrupted streaming. Our main contribution is to model the playout buer in the user device with a queuing approach,

and to provide a mathematical formulation for the idle periods of this buer, which relate to the bytes downloaded from the cellular infrastructure. We also solve the resulting

content allocation problem, and perform trace-based simulations to nally show that up to 50% of the original trac could be ooaded from the main infrastructure.