MOOSE: Enabling Massively Parallel Multiphysics Simulation (1911.04488v1)

Abstract: Harnessing modern parallel computing resources to achieve complex multi-physics simulations is a daunting task. The Multiphysics Object Oriented Simulation Environment (MOOSE) aims to enable such development by providing simplified interfaces for specification of partial differential equations, boundary conditions, material properties, and all aspects of a simulation without the need to consider the parallel, adaptive, nonlinear, finite-element solve that is handled internally. Through the use of interfaces and inheritance, each portion of a simulation becomes reusable and composable in a manner that allows disparate research groups to share code and create an ecosystem of growing capability that lowers the barrier for the creation of multiphysics simulation codes. Included within the framework is a unique capability for building multiscale, multiphysics simulations through simultaneous execution of multiple sub-applications with data transfers between the scales. Other capabilities include automatic differentiation, scaling to a large number of processors, hybrid parallelism, and mesh adaptivity. To date, MOOSE-based applications have been created in areas of science and engineering such as nuclear physics, geothermal science, magneto-hydrodynamics, seismic events, compressible and incompressible fluid flow, microstructure evolution, and advanced manufacturing processes.

• The paper demonstrates MOOSE’s capability to streamline the development of complex multiphysics simulations using a modular finite element method architecture.
• The paper details an innovative framework design that leverages Kernel and BoundaryCondition systems, along with in-situ postprocessing, to simplify PDE specification.
• The paper validates MOOSE’s performance through extensive scaling studies on supercomputers, achieving efficient computation on tens of thousands of processor cores.

MOOSE: Enabling Massively Parallel Multiphysics Simulation

The paper presents MOOSE (Multiphysics Object Oriented Simulation Environment), an advanced computational framework designed to facilitate the development and execution of multiphysics simulations across a wide range of scientific and engineering domains. MOOSE offers a highly modular architecture, enabling efficient and scalable solutions to complex problems by leveraging parallel computing resources without imposing the burden of intricate parallel code management on the user.

Core Architecture and Capabilities

MOOSE adopts a robust finite element method (FEM) foundation, built upon well-established libraries such as libMesh and PETSc, ensuring numerical stability and performance. Its architecture revolves around a system of extensible classes and interfaces that permit users to define partial differential equations (PDEs), boundary conditions, and material properties in a modular and reusable fashion. This modularity allows for significant code reuse and collaborative development across research teams.

One of the notable features of MOOSE is its Kernel and BoundaryCondition systems, which enable users to define the weak form of PDEs with minimal code. This abstraction simplifies the specification of complex models spanning multiple dimensions and physical domains. Furthermore, the framework's material property system enhances flexibility by allowing nonlinear property definitions and automatic Jacobian updates through its automatic differentiation (AD) capability.

The in-situ postprocessing functionality within MOOSE represents another advanced feature. It permits computations traditionally performed post-simulation to be carried out simultaneously with the main simulation, offering enhanced efficiency and the possibility of feedback into ongoing calculations.

Parallelism and Scalability

MOOSE's parallel architecture employs a hybrid model utilizing both MPI and shared-memory threading, providing broad scalability across vast computational resources. It demonstrates scalability over tens of thousands of processor cores, supported by advanced mesh handling strategies, such as pre-splitting to optimize memory usage and computational load distribution. This capability is crucial for applications involving large-scale simulations requiring billions of unknowns.

The paper details a scaling paper on the Theta supercomputer at Argonne National Laboratory, achieving efficient performance with up to 32,768 processor cores, indicating MOOSE's capability to handle demanding computations in various scientific fields effectively.

Multiscale and Multiphysics Integration

Through the MultiApp system, MOOSE facilitates seamless integration of independently developed physics simulations, enabling multiscale modeling. This ability is exemplified by coupling engineering scale problems with microstructural evolution simulations, highlighting its adaptability to diverse problem scales without additional custom coding to link applications.

Practical Implications and Future Directions

MOOSE's workflow encourages best practices in scientific software development, promoting code reuse, modularity, and extensibility, which can significantly enhance research productivity and software reliability. As simulation demands become increasingly complex, frameworks like MOOSE that lower the entry barrier to parallel multiphysics modeling become indispensable.

The extensive testing infrastructure, robust visualization tools, and comprehensive documentation contribute to the platform's usability and adoption. MOOSE's active development community and its open-source nature suggest a trajectory of continued enhancement and adaptation to emerging computational needs.

In future iterations, MOOSE may further extend its reach and capabilities, potentially incorporating more sophisticated machine learning integrations or expanding support for emerging hardware architectures. Such developments could further solidify its role as a vital tool in the scientific computing landscape.

Overall, MOOSE stands out as a highly capable framework designed to streamline the development of sophisticated multiphysics simulation tools, enabling researchers to focus on scientific inquiry rather than computational intricacies.