CRII: ACI: Efficient Radiative Heat Transfer Modeling In Large-Scale Combustion Systems

热辐射是一个重要的传热过程，但对大规模热系统的了解还不够充分。在大规模火灾中，辐射传热占向周围环境总传热的70%以上。燃烧系统产生的污染物，如颗粒物质、氮氧化物和硫氧化物（酸雨的主要贡献者），对辐射传热的热效应高度敏感。为了正确预测火灾传播和污染物排放，指导电厂改造，需要对大型燃烧系统进行高保真的辐射建模。然而，高保真辐射模型昂贵的计算成本，它们密集的内存需求和糟糕的缩放性能传统上阻碍了它们在玩具或小规模问题之外的应用。现代高性能计算系统已经发展到可以利用大规模并行性的阶段，但每个核心的内存正在减少。因此，需要新的热辐射预测建模和并行策略来利用当前和未来网络基础设施的力量。为了促进对热辐射过程的理解，并使预测模型能够应用于实际工程系统，本CRII项目旨在优化现代异构多核高性能计算系统的高保真辐射模型的求解算法和并行化策略。因此，这项研究与美国国家科学基金会促进科学进步和促进国家健康、繁荣和福利的使命是一致的。该项目的总体目标是利用现代网络基础设施，打破将高保真辐射模型应用于实际大规模系统的障碍。这样可以更好地预测计算边界上的热流密度和污染物排放，从而减少火灾损失，缓解污染对环境的担忧。具体而言，该项目侧重于利用多集成核（MIC）协处理器提供的混合计算环境，增强基于蒙特卡罗的高保真辐射模型的并行性。根据建议，高保真辐射模型将耦合到一个开源的火灾模拟器，并将根据充分记录的实验数据进行验证。通过识别不同物理过程的不同时间尺度，首先对求解算法进行优化，以提高所提代码的整体效率。然后提出了消息传递接口（MPI）和OpenMP的混合并行性，以实现所需的“解决时间”的减少。最后，将通过大规模的火灾模拟来验证所开发的火灾辐射规范的准确性和效率。

Thermal radiation is an important but less adequately understood heat transfer process for large-scale thermal systems. Radiative heat transfer accounts for more than 70% of the total heat transfer to the ambient environment in large-scale fires. Pollutants that are produced by combustion systems, such as particulate matters, NOx and SOx (main contributor to acid rain), are highly sensitive to the thermal effects of radiative heat transfer. To correctly predict fire propagation and pollutant emission, and to guide power plant retrofit, high-fidelity radiation modeling for large-scale combustion systems is needed. However, the expensive computational cost of high-fidelity radiation models, their intensive memory requirements, and poor scaling performances have traditionally prevented their applications beyond toy or small-scale problems. Modern high performance computing systems have evolved to a stage where massive parallelism can be harnessed but memory-per-core is decreasing. Therefore, new modeling and parallelism strategies for thermal radiation prediction are required to leverage the power of current and future cyber-infrastructure. To advance the understanding of the thermal radiation processes, and to enable the application of predictive models to practical engineering systems, this CRII project aims at optimizing the solution algorithms and parallelism strategies of high-fidelity radiation models for the modern heterogeneous many-core high performance computing systems. Therefore, this research aligns with the NSF mission to promote the progress of science and to advance the national health, prosperity and welfare.The overarching goal of this project is to break the barrier of applying high-fidelity radiation models to practical large-scale systems, utilizing modern cyber-infrastructure. As a result, the heat flux on the computational boundaries as well as pollutant emissions can be better predicted, which can reduce the fire loss and alleviate the environmental concerns with pollutions. Specifically, the project focuses on enhancing the parallelism of a Monte Carlo based high-fidelity radiation model, using the hybrid computing environment provided by the many-integrated-core (MIC) co-processors. As proposed, the high-fidelity radiation model will be coupled to an open-source fire simulator, and will be validated against well-documented experimental data. By identifying the disparate time scales of different physical processes, solution algorithm is first optimized to enhance the overall efficiency of the proposed code. Hybrid parallelism with message passing interface (MPI) and OpenMP is then proposed to achieve the desired reduction in the "time to solution." Finally, the accuracy and efficiency of the developed fire-radiation code will be demonstrated through a large-scale fire simulation.