摘要
对流应用系统进行吞吐量分析需要将周期静态顺序调度建模到数据流图中,吞吐量分析效率依赖于数据流图的规模及建模时间。为了提高吞吐量分析效率,提出基于同构同步数据流图的调度感知同步数据流模型及相应建模方法。通过利用应用模型结构特征及周期静态顺序调度,可减少模型中的任务、边和初始符号数目;可以使用已有分析方法对模型进行吞吐量分析。实验结果表明,所提建模方法优于已有方法,可有效提高吞吐量分析效率。
To analyze the throughput of streaming application systems, it is necessary to model the periodic static order schedule into the synchronous dataflow graph. The throughput analysis efficiency depends on the size of the dataflow graph and the modeling time. To improve the throughput analysis efficiency, a schedule-aware dataflow model and its modeling method were proposed on the basis of the homogeneous synchronous dataflow graph. By exploiting the structure of the application model and the periodic static order schedule, the task number, edge number and initial token number were reduced. Besides, the throughput of the model can be analyzed using available analyzing methods. Experimental results show that the proposed modeling method outperforms the available methods, with the throughput analysis being optimized efficiently.
作者
唐麒
吴尚峰
施峻武
魏急波
TANG Qi WU Shangfeng SHI Junwu WEI Jibo(College of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073 , China)
出处
《国防科技大学学报》
EI
CAS
CSCD
北大核心
2017年第2期128-133,共6页
Journal of National University of Defense Technology
基金
国家自然科学基金资助项目(61471376)
关键词
同步数据流图
调度感知
多处理器
状态空间
synchronous dataflow graph
schedule-aware
multiprocessor
state space
