An Accelerated Edge Computing with a Container and Its Orchestration

Abstract

Current mobile applications are becoming more compute-intensive so that they should be offloaded to central, fog, or multi-access edge cloud (MEC). Even if the application is offloaded to a cloud and there exists more computation power compared to mobile devices, only using CPU might not be sufficient to serve a service in real-time. This makes adopting hardware accelerator such as GPU and FPGA more attractive. However, there is no cloud framework supporting heterogeneous hardware accelerators. In this paper, we propose a container-based cloud framework that makes a container possible to use heterogeneous accelerators by automatically providing pass-through between the container and accelerators. Our framework, which is based on Docker and Kubernetes, reduces container size compared to the container with manually configured pass-through and makes the container to be allocated to a proper node where the accelerator is installed. In the evaluation section, we demonstrate the efficiency of our framework in terms of the container size and the container allocation overhead.