基于负载均衡的CPU-GPU异构计算平台任务调度策略

Task Scheduling Strategy of CPU-GPU Heterogeneous Computing Platform Based on Load Balancing

针对中央处理单元-图形处理单元(central processing unit-graphics processing unit,CPU-GPU)异构计算系统中,CPU和GPU负载不均导致系统性能降低的问题,提出了一种基于队列的混合调度策略.该策略通过探测获得CPU和GPU处理指定任务的计算能力,将计算任务按照探测比例分配给CPU和GPU;将并行任务存入双向队列,以降低调度带来的额外开销.结果表明,使用该策略的基准测试程序系统性能平均提升了28.07%.总体而言,该调度策略能够缩短CPU与GPU完成各自计算任务后的等待时间,有效平衡系统CPU与GPU之间的负载,提升系统性能.

In central processing unit-graphics processing unit(CPU-GPU)heterogeneous system,the uneven performance of the CPU and GPU caused the system performance to decrease.A hybrid scheduling strategy based on queues was proposed to solve the problem.The computing power of the CPU and GPU was detected to process specified tasks,and computing tasks were allocated to the CPU and GPU according to the perception ratio.The tasks were stored in a bidirectional queue to reduce the additional overhead brought by scheduling.Results show that the system performance of the benchmark test program is improved by using this strategy by an average of 28.07%.Overall,the scheduling strategy can reduce the waiting time after the CPU and GPU complete their respective computing tasks,balance the load between the system CPU and GPU,and improve the system performance.