Research And Implementation Of Task Scheduling And Placement In CPU-FPGA Heterogeneous System

With the development of information technology such as big data and cloud computing and the growth of data scale,the demand for computing power of the system is gradually increasing.The traditional general CPU has been unable to meet the computing demand of complex algorithm and large-scale data.CPU-FPGA heterogeneous system is gradually becoming one of the main research directions of high-performance computing.FPGA's unique reconfigurable features can provide different functions for the system,but at the same time,task scheduling and task placement in the system are facing many challenges.First of all,resource fragmentation is still serious and the diversity of reconfigurable resources is not considered in traditional task placement strategy which result in the underutilization of resources.Secondly,time consumption caused by reconfiguration and resource contention among multiple tasks affect the execution efficiency of the system to a certain extent.This paper focuses on the task scheduling and placement in CPU-FPGA heterogeneous system.Firstly,this paper proposes a multi-resource task placement strategy in single FPGA which considers the requirements of tasks for different hardware resources in a fine-grained way and synthesizes the time of tasks for different resources.It selects the appropriate placement location for tasks by using three-dimensional space-time model which makes full use of hardware resources and reduces the degree of fragmentation of resources.Secondly,in order to improve the performance of multi-FPGA system,this paper proposes a two-level task scheduling strategy.In the task level scheduling stage,the similarity of sub task type and resource demand is considered comprehensively to select the appropriate FPGA for tasks.It can reduce some reconfigurable time consumption and resource contention.In the sub task level scheduling stage,based on the task placement algorithm,the genetic algorithm and the two terminal ant colony optimization algorithm are used to find the optimal scheduling sequence and placement location for the sub task set.In this way,the hardware resources of FPGA are fully utilized and the total completion time of the task set are reduced and improve the execution efficiency of the system.Finally,TGFF is used to generate topology tasks and the task scheduling and placement methods proposed in this paper are compared with the traditional methods.The experimental results show that task placement method described in this paper can effectively reduce resource fragmentation and improve the average utilization of resources.In addition,in the multi-FPGA system,the proposed two-level task scheduling method makes more balanced use of hardware resources and reduces the total execution time of tasks than the traditional methods.