Near-optimal content replication for interactive multiview video streaming

In interactive multiview video streaming (IMVS), a client can watch the multiview video and interact with it by switching to different viewing angles at frozen moments. To provide scalable IMVS services, a content provider often deploys distributed content servers with heterogeneous storage capacities in order to collaboratively replicate video segments for their clients. IMVS presents a new challenge in content replication: which video segments to replicate to support interactive view-switching during an IMVS session. In this paper, we first formulate the content replication problem as an integer linear programming (ILP) problem, which is proven to be NP-hard. We then propose a fast algorithm to solve it with bounded approximation error. In this algorithm, we first solve the ILP problem as a relaxed LP problem, and then heuristically round the resulting fractional LP solution to integer towards ILP feasibility. Simulation results show that our replication strategy can substantially reduce access costs compared to a commonly used replication scheme and a state-of-the-art replication scheme.