Wireless Video Content Delivery through Distributed Caching and Peer-to-Peer Gossiping Negin Golrezaei, Alexandros G. Dimakis, Andreas F. Molisch, Giuseppe Caire Dept. of Electrical Eng. University of Southern California emails: {golrezae,dimakis,molisch,caire}@usc.edu Abstract—We present a novel approach to handle the ongoing explosive increase in the demand for video content in wireless mo- bile devices. We show how distributed caching and collaboration between users and femtocell-like base stations without high-speed backhaul, which we call helpers, can greatly improve throughput without suffering from the backhaul bottleneck problem common to femtocells. We also investigate the role of collaboration among users - a process that can be interpreted as the mobile devices playing the role of helpers also. This approach allows an improvement in the video throughput without deployment of any additional infrastructure. The efficiency of the caching approach depends on two key system properties: (i) the configuration of the ”effective” distributed cache, i.e., which user can connect to which helpers and (ii) the popularity distribution of the video files. As a function of these properties, we consider the wireless distributed caching problem, i.e., which files should be cached by which helpers. For the uncoded helper problem, a strictly optimum solution of this problem is NP-hard, but good approximation algorithms can be found. Furthermore if distributed coding is used for the caching, exact optimization can be done with polynomial complexity. For the configuration with mobiles-as-helpers problem, we adopt a simplified model wherein users are grouped into clusters within which communications is possible. A key question is the choice of the cluster dimension (collaboration distance), trading off spectral reuse with the probability of finding the desired video within the collaboration distance. Numerical optimization as well as general scaling laws are described. Simulations show that our approach can improve the data throughput by as much as 400 - 500% through addition of helpers, and more than an order of magnitude through the device- to-device communications. I. I NTRODUCTION Over the past decade, the proliferation of smartphones, tablets, and other wireless devices has led to an explosive growth of wireless data traffic. While initial growth was mainly driven by web-browsing and associated applications, emphasis is now shifting to wireless delivery of video content, which is expected to lead to an increase in wireless data traffic by two orders of magnitude in the next few years. [1]. Current systems already have close-to-optimum technology for their physical link between transmitter and receiver. Thus, the most promising ways to achieve such an enormous increase in area spectral efficiency and thus data throughput, of such tremendous scale is a decrease in cell size, essentially bringing the content closer to the users. The common method of realizing this is the deployment of small base stations that enable high-density spatial reuse of communication resources [2]. Such pico- and femto-cell networks, which are usually combined with macrocells into a heterogeneous network, are receiving a lot of attention in the recent literature, (see e.g. [3] and references therein). However, a major practical obstacle to this approach is the requirement for high-speed backhaul to the cellular operator network [3]. In this paper we survey our on-going work [4]–[6] of using distributed storage and collaboration to address the backhaul bottleneck problem. Our main idea is to replace the femto-base stations by small base stations that have a (possibly wireless) low-bandwidth backhaul link but high storage capacity. These stations, which we henceforth call caching helpers, or simply helpers, form a wireless distributed caching infrastructure. Further in our on-going work [6] we investigate using other mobile users as helpers: popular content is cached and through base station assisted device-to-device (BSA-D2D) gossiping is delivered to new users who request it. Distributed storage can be exploited at different levels and fueled by short-range high- throughput local connections. Predictability of the popular content and the massive distributed storage capacity yield that, with high probability, the video that any given user wishes at any given time is already in someone’s cache in the same macro-cell. A novel wireless BSA-D2D wireless architecture takes care of delivering the desired content to the users within one short-distance hop through local D2D links. This architecture can lead to significant gains if there is sufficient content re-use, i.e., a significant fraction of the requests correspond to few popular files. We now proceed to give an overview of our novel research contributions: 1) Femtocaching: How to optimize the file placement in helpers to maximize the probability of finding popular files locally in a physically distributed cache [4]. 2) Coded Femtocaching: How the splitting of popular con- tent into blocks and the encoding of these blocks allows the relaxation of an NP-complete problem into a convex optimization of linear program [5]. 3) Device-to-Device Collaboration: How to turn other users into helpers by distributed caching of popular content in other mobile devices. The problem of finding the optimal collaboration distance and the scaling laws that can be achieved through this architecture [6], [7].