基于异构多核平台的同步数据流图帕累托优化与调度

发布时间：2018-05-22 16:19

  本文选题：同步数据流图 + 异构多核平台　； 参考：《计算机科学》2015年11期

【摘要】：同步数据流图被广泛用于多媒体和数字信号处理程序等流应用程序的建模。流应用程序须达到一定吞吐量才能流畅运行,利用异构多核处理器来进一步提高流应用程序的吞吐量已经成为当今嵌入式系统的发展趋势,但是提高吞吐量往往伴随着能耗的增加。为了解决这个问题,基于异构多核平台的同步数据流图系统模型,给出了求解所有能耗和吞吐量的帕累托优化点及其相应静态调度的方法。首先将系统模型转换为时间自动机网络,并将分析目标转换为时序逻辑公式;再使用实时模型检测工具UPPAAL寻找解决方案;最后对UPPAAL返回的结果进行分析,找出满足要求的调度。由于模型检测方法可对问题空间进行穷尽搜索,该方法得到的结果是精确的。该方法可帮助设计者在系统开发早期了解系统能耗和吞吐量的量化关系,有利于缩短系统的开发周期,降低开发成本。
[Abstract]:Synchronous data flow diagrams are widely used to model multimedia and digital signal processing applications. Streaming applications must reach a certain throughput to run smoothly. Using heterogeneous multi-core processors to further improve the throughput of streaming applications has become the development trend of embedded systems. But increasing throughput is often accompanied by an increase in energy consumption. In order to solve this problem, based on the synchronous data flow graph system model of heterogeneous multi-core platform, a method for solving Pareto optimization points of all energy consumption and throughput and its corresponding static scheduling is presented. Firstly, the system model is transformed into a time-automaton network, and the analysis object is converted into a temporal logic formula; then a real-time model checking tool UPPAAL is used to find a solution; finally, the results returned by UPPAAL are analyzed to find out the scheduling that meets the requirements. Because the model detection method can search the problem space completely, the result obtained by this method is accurate. This method can help designers understand the quantitative relationship between system energy consumption and throughput in the early stage of system development, shorten the development cycle and reduce the development cost.
【作者单位】： 苏州大学计算机科学与技术学院;中国科学院软件研究所计算机科学国家重点实验室;
【基金】：973课题(2014CB340701) 江苏省自然科学基金(BK2011281) 苏州市应用基础研究计划(SYG201241) 江苏省研究生科研创新计划(KYLX_1247)资助
【分类号】：TP332;TP311.11
，

本文编号：1922784