Scalable Design Space Exploration via Answer Set Programming

本研究项目的目标是在探索速度和适用性方面持续改善嵌入式和网络物理系统的自动设计空间探索。设计空间探索的任务是为这些系统的实现确定不同的最佳解决方案。为此，设计并研究了基于答案集规划的嵌入式系统综合与优化方法。在探索过程中，将以复杂的静态和动态设计决策同时考虑的方式扩展应用范围。由此产生的设计空间大大增加，这往往阻碍了最先进的方法的有效探索。通过直接结合非单调约束检查来提高所研究的勘探方法的有效性，非单调约束检查通常源于嵌入式系统设计中的延迟、吞吐量和功耗要求。我们研究(1)新颖的设计空间探索方法如何从答案集规划的新发展中受益，以及(2)答案集规划领域的新方法，这些方法能够有效地进行设计空间探索。后者包括多目标优化的集成和其他理论求解的结合。在第二个项目阶段，我们(1)将重点从设计空间探索中的选择性方法转移到生成系统级方法，即在设计空间探索过程中，我们从组件库中构建针对工作负载场景进行优化的计算平台，而不是从固定的应用程序规范和平台模板开始。为此，我们正在利用答案集编程的多功能求解能力。此外，我们(2)还将基于答案集编程的决策集成到嵌入式计算平台本身，以支持由嵌入式映射器做出的动态设计决策。最后，我们(3)将嵌入式映射器整合到新型生成设计空间探索的背景理论中，以允许对设计和运行时决策之间的交互进行协调评估。从而保证了所生成的计算平台在给定工作负荷下的质量特征，包括其动态决策。通过研究基于答案集规划的设计空间探索，不仅在设计自动化领域，而且在答案集规划领域都有望取得科学成果。

The goal of this research project is to sustainably improve the automatic design space exploration of embedded and cyber-physical systems with respect to exploration speed and applicability. Design space exploration is the task of identifying different optimal solutions for the implementation of these systems. For this purpose, novel embedded systems synthesis and optimization methods based on answer set programming are designed and studied. The area of applicability will be extended in a way that both, complex static but also dynamic design decisions, are considered during exploration. The resulting design space is enormously increased, which often prevents state-of-the-art methods from effective exploration. The effectiveness of the studied exploration methods is increased by directly incorporating non-monotonic constraint checking, which often stems from latency, throughput, and power consumption requirements in embedded systems design. We study (1) how novel design space exploration methods can profit from new developments in answer set programming as well as (2) novel approaches in the domain of answer set programming, which enable effective design space exploration. The latter includes the integration of multi-objective optimization and the incorporation of other theory solvers. In the second project phase, we (1) shift our focus from selective methods in design space exploration to generative system-level methods, i.e. instead of starting with a fixed specification of applications and platform templates, we construct computing platforms, which are optimized for workload scenarios, from a component library during design space exploration. For this purpose, we are exploiting the versatile solving capabilities of answer set programming. Moreover, we (2) also integrate decision-making based on answer set programming into the embedded computing platform itself in order to support dynamic design decisions that are made by an embedded mapper. Finally, we (3) incorporate the embedded mapper into a background theory of the novel generative design space exploration to allow for a coordinated assessment of the interaction between design and run time decisions. As a consequence, guarantees on the quality characteristics for the generated computing platform including its dynamic decision making under a given workload can be given. By studying design space exploration on the basis of answer set programming, scientific findings are not only expected in the domain of design automation but also in the domain of answer set programming.