FPGA virtualization deployment based on Docker container technology

Abstract

FPGA (Field Programmable Gate Array) has special advantages in low latency, low power consumption and reconfigurable hardware as compared with CPU and GPU. It is deployed as heterogeneous accelerators on a large scale in cloud computing centers and research laboratories. Usually., developers operate directly on the physical machines and use the FPGA card exclusively. This leads to a bad utilization of FPGA resources. This paper proposes to virtualize and deploy FPGA resources as lightweight containers by using Docker and Kubernetes together. The Docker containers abstract FPGA resources and provide an isolated runtime environment for each application. The Kubernetes performs unified and automated scaling and scheduling of virtualized FPGA containers. Thus, FPGA resources can be safely and effectively shared among multiple local and remote applications. Experiments show that the utilization of FPGA resources has been improved.