An architecture design of GPU-accelerated VoD streaming servers with network coding

Abstract

Graphics processing unit (GPU) has evolved into a general-purpose computing platform. Inspired by the GPU technology advantage, this paper concerns the design and performance evaluation of practical GPU-accelerated server architecture for Video-on-Demand (VoD) services with network coding. Following the proposal of an optimized network coding algorithm based on parallel threads on GPU, a GPU-Accelerated Server (GAS) for VoD streaming is designed to leverage the workload between GPU and CPU and thus improve the performance of the VoD server. Extensive real-world experimental results have proved that compared with the approaches with network coding performed only on CPU or GPU, the proposed GAS architecture is more advantageous in serving capacity, response time, and CPU usage. Our study has investigated a way of designing high performance VoD streaming servers with network coding and GPU-acceleration incorporated.