摘要
CPU-GPU异构多核系统对计算密集型的应用加速效果显著而得到广泛应用,但易出现负载均衡问题。针对此问题,提出了一种CPU-GPU异构多核系统的动态任务调度算法。该算法充分利用CPU的线程资源和GPU的计算资源,准确测量CPU和GPU的计算能力,从而动态调整分配到CPU和GPU上的数据块大小,减小负载的总执行时间,提高系统加速比。实验结果表明,该算法使得系统加速比提高34%~103%。
CPU-GPU heterogeneous multi-core system has been widely applied because of its acceleration effects for compute- intensive applications. However, the problem of workload imbalance is serious. Therefore, this paper proposed a dynamic task scheduling algorithm （DTSA） based on CPU-GPU heterogeneous multi-core system. In order to guarantee that all cores were doing useful work, it made full use of CPU and GPU. Furthermore, it could accurately measure the computational power of GPUs and CPUs respectively, dynamically adjusted the size of data blocks to be executed on CPUs and GPUs, and finally re- duced the total executing time of workloads and increased the system speedup. According to the results of experiments by using this algorithm, the system speedup increases by 34% - 103%.
出处
《计算机应用研究》
CSCD
北大核心
2016年第11期3315-3319,共5页
Application Research of Computers
基金
上海市自然科学基金资助项目(15ZR1428600)
计算机体系结构国家重点实验室开放资助项目(CARCH201206)
上海市浦江人才计划资助项目(16PJ1407600)
关键词
动态调度
负载均衡
自适应分配
异构计算
dynamic scheduling
workload balance
adaptive allocation
heterogeneous computing
