Research On Multi-task Scheduling Technology Based On CPU+GPU Heterogeneous Signal Processing Platforms

With the increasing real-time requirements of high-speed signal processing,the highperformance processing platform based on CPU+GPU architecture has been paid more and more attention in the field of signal processing.In order to give full play to the efficiency of CPU and GPU computing resources in the platform,the task scheduling algorithm needs to match the hardware characteristics of the platform.However,for CPU+GPU heterogeneous signal processing platform,due to the particularity of CPU+GPU architecture,the traditional scheduling algorithm can't give full play to its advantages,and also has adverse effects on the performance of the platform.According to the characteristics of hardware resources of CPU+GPU heterogeneous signal processing platform,the task scheduling decision and method suitable for the platform are improved,and the effective management and utilization of a variety of parallel computing resources on the platform are realized.The main work of this paper is as follows:Firstly,aiming at the problem that the existing scheduling algorithms can't allocate computing tasks according to the hardware characteristics of CPU+GPU heterogeneous signal processing platform,a load deployment decision(LDD)algorithm based on CPU + GPU heterogeneous signal processing platform is proposed.The algorithm is improved on the basis of HEFT(heterogeneous early finish time)algorithm.It introduces the load deployment decision,and divides the tasks that are not suitable for GPU to CPU in advance,so as to give full play to the powerful parallel computing advantage of GPU,avoid the GPU processing tasks with small amount of computation,and let CPU participate in parallel computing.The simulation results show that LDD algorithm can effectively improve the utilization of multiprocessors on CPU+GPU heterogeneous signal processing platform,standardize the execution order of tasks,shorten the overall scheduling time,balance the load,and achieve the purpose of improving scheduling.Secondly,aiming at the problem that the existing scheduling algorithms can't solve the transmission constraints between different processors on heterogeneous signal processing platforms,this paper proposes the longest path list scheduling(LPLS)algorithm for heterogeneous signal processing platforms.In the task priority stage,the priority is calculated based on the longest path list,so that the tasks on the most time-consuming path are scheduled first;In the processor selection stage,the processor is selected according to the principle of minimizing the sum of the earliest completion time of the task and the longest path from the follow-up task to the exit task,so that the overall processing time of the heterogeneous platform is smaller.The simulation results show that compared with the classic HEFT algorithm,LPLS algorithm is a more load balanced algorithm with shorter scheduling length and higher efficiency.Thirdly,in order to test the scheduling effect of LDD algorithm and LPLS algorithm,the two scheduling algorithms are fused on CPU+GPU heterogeneous signal processing platform to form a multi-core parallel scheduling(MCPS)algorithm suitable for CPU+GPU heterogeneous signal processing platform.By accelerating the scheduling of the PTS-MUSIC(parallel MUSIC algorithm with task schedule strategy)algorithm,it is verified that MCPS algorithm not only improves the utilization of processor parallel resources on CPU+GPU heterogeneous signal processing platform,but also accelerates the processing of signal data,and greatly shortens the processing time of the PTS-MUSIC algorithm.