基于异构多核平台的同步数据流图帕累托优化与调度

Pareto Optimization and Scheduling of Synchronous Dataflow Graphs on Heterogeneous Multicore Platform

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。

Synchronous dataflow graphs （SDFGs） are widely used to model streaming applications such as multimedia and digital signal processing applications. Streaming applications are usually required to reach a high throughput to guarantee smooth running. Using heterogeneous multicore processors to improve the throughput of streaming applica- tions has become a feasible solution. However, a higher throughput is usually achieved with the increase of energy con- sumption. We presented a method to explore the Pareto space of energy consumption and throughput and find the schedule of each Pareto point. A system model includes an SDFG description for the application and a heterogeneous multicore platform. The system model is transformed to a network of timed automata （NTA） and the optimization goals are formalized as temporal logic formulae. The NTA and formulae are then used as the input of the real time model checking tool UPPAAL. The Pareto points and corresponding schedules by the trace provided by UPPAAL can be ob- tained. Our method is exact, which is benefited from the exhaustive search nature of model checking technique. The method can he used to help designers to understand the quantitative relationship between energy consumption and throughput in the design period, shortening development cycles and reducing costs of system developments.