"HiLeMMS": High-Level Mesoscale Modelling System

中尺度模型在解决涉及多尺度和多物理流体流动的广泛的工程和科学挑战方面发挥着非常重要的作用。为了准确地模拟这些创新的流动，必须有一个计算机建模系统，能够有效地管理复杂的几何和数据结构，并充分利用新兴硬件架构带来的高性能计算能力。然而，这些被证明是两大障碍：虽然没有适用于所有流的单一几何描述和数据结构方法，以及需要不同编程技术的竞争计算平台，但对于单个研究小组来说，开发这样一个可以在不同几何描述和不同硬件平台之间轻松切换的系统是非常耗时的，如果不是不可能的话。为了应对这一挑战，我们建议在现有基础设施的基础上开发一个高级别中尺度模拟系统(HiLeMMS)。已经有利用高级抽象方法来隐藏并行性的细节并处理几何描述和数据结构的单一方法的成功努力，例如来自牛津大学的用于多块应用的结构化网格解算器的牛津并行库(OPS)和来自劳伦斯伯克利国家实验室的用于自适应网格精化的Chombo库。重要的是，这些库将支持主要的新兴硬件架构。因此，我们将通过在这些库之上构建专门用于格子Boltzmann方法的高级抽象层来构建HiLeMMS。通过这种方式，HiLeMMS将隐藏和自动化利用现有库所需的编程，并加快应用程序开发。应用程序开发人员只需要编写一次代码，然后就可以灵活地切换库，以最小的努力获得科学问题所需的能力。DL_Meso包(DL_Meso_LBE)中的格子Boltzmann代码将使用HiLeMMS进行重新设计，以便代码能够管理不同的几何和数据结构，以及在新兴硬件平台上高效执行。得益于高层框架，我们将进一步开发一些新功能到该套装中，扩展对非连续气体流动、流体-结构相互作用和多相流的模拟能力。

Mesoscale modelling plays a very important role in addressing a broad range of engineering and science challenges involving multiscale and multiphysics fluid flows. To accurately simulate these innovative flows, it is necessary to have a computer modelling system capable of efficiently managing complex geometries and data structures, and harnessing fully high performance computing power brought by emerging hardware architectures. However, these prove to be two major obstacles: while there is no single method of geometry description and data structure suitable for all flows, as well as competing computing platforms requiring different programming techniques, it is very time-consuming if not impossible for a single research group to develop such a system that can easily switch between different geometry descriptions and different hardware platforms. To tackle this challenge, we propose to develop a high-level mesoscale modelling system (HiLeMMS) on the basis of existing infrastructures. There have been successful efforts exploiting a high-level abstraction approach to hide the details of parallelism and handling a single method of geometry description and data structure, such as the Oxford Parallel Library for Structured-mesh solvers (OPS) for multi-block applications from Oxford University and the Chombo library for adaptive mesh refinement from the Lawrence Berkeley National Laboratory. Importantly, these libraries will support major emerging hardware architectures. Therefore, we will build HiLeMMS by constructing a high-level abstraction layer specifically for the lattice Boltzmann method on top of these libraries. In this way, HiLeMMS will hide and automate the programming required for utilising existing libraries, and speed up application development. The application developer will only need to write code once, and then will enjoy the flexibility of switching libraries with minimal effort to obtain the capability required for the scientific problem.The lattice Boltzmann code in the DL_MESO package (DL_MESO_LBE) will be re-engineered using HiLeMMS so that the code can have the capability of managing different geometry and data structures as well as efficient execution on emerging hardware platforms. Benefiting from the high-level framework, we will further develop a number of new functionalities into the package, extending the capabilities to model non-continuum gas flows, fluid-structure interactions and multiphase flows.