A Cooperative Method of Task Scheduling based on FPGA Cloud Platform

A cloud platform based on FPGA acceleration cards cluster and servers cluster is proposed in this paper. Multi-type boards in the cloud platform could provide multi-type application acceleration services to meet the needs of different tenants. Therefore, we propose to design an optimal task scheduling and service scheme to maximize tenants satisfaction and minimize operators benefits loss. Firstly, the problem of task scheduling is considered as a `max-min' joint optimization problem. Then it is transformed into a bi-level optimization problem with the lower as tenants' satisfaction maximization problem and the upper as FPGA cloud platform operators benefits loss minimization problem. Finally, because the lower problem is convex, the MATSuMoTo algorithm is used to obtain the best task scheduling strategy of the cloud platform.