(DURIP-FY22) A MULTI-ARCHITECTURE HARDWARE COMPUTING CLUSTER FOR THE DEVELOPMENT AND EFFICIENT IMPLEMENTATION OF A VARIETY OF ROBUST AND SCALABLE NUMERICAL ALGORITHMS
Abstract
The development of efficient, robust, scalable, and accurate numerical algorithms suitable for modern high performance computing architectures has become a major ingredient needed for simulation of complex problems such as fluid simulations, combustion, design of materials, and many others. The continual challenge in high performance scientific computing is the need for midrange cutting-edge university-based equipment to enable the application-driven development of innovative numerical algorithms to simulate complex multiscale physical problems. Furthermore, researchers find themselves in need of multiple types of hardware to develop methods for solution of problems that require different modes- an intensive offline simulation mode and an efficient online reconstruction of a solution in a low-dimensional space approximating the high fidelity solution manifold. This proposal aims to directly meet these challenges by acquiring a new multi-architecture hardware computing cluster, that will become a shared campus research instrument for an inter- and multi-disciplinary group of mathematicians and computational scientists and engineers, and their research groups. The proposed cluster will contain 72 multi-core x86 compute nodes, 5 multi-GPU compute nodes, 2 ARM-based compute nodes and a petabyte of attached storage, connected over a high-speed network. This computing cluster will foster the development and testing of novel mathematical methods suitable for large scale parallel scientific computing and data science applications. These novel mathematical algorithms will be designed for cutting-edge scalable performance, distributed-memory parallel, and GPU-based numerical simulations, as well as novel scientific data science and data analytics approaches. In particular, eight research projects will be enabled by this cluster- (1) Developing a numerical analysis framework for developing energy and power efficient algorithms; (2) Development of reduced order algorithms for simulation of quantum materials and computation of magic angles ; (3) Developing quantum mechanical (QM) simulations for accelerating Materials Discovery for Energy Storage Applications; (4) Developing PDE solvers and machine learning and data analytics techniques for black hole simulations; (5) Machine learning techniques for design of metamaterials for enhanced source localization; (6) Reduced basis approaches for robust material design as a topology optimization under uncertainty problem; (7) Development of fuel spray-wall interaction models for predictive simulations of turbulent combustion flow; (8) Partition of unity multivariate approximation for the volume of fluid method. These projects focus on the development of innovative computational mathematics and data science algorithms that are of interest to the AFOSR, including development of novel and efficient algorithms; mathematical methods for model complexity reduction; data analytics approaches including machine learning and similar approaches; development of novel materials including metamaterials, multi-phase architected materials, quantum materials, and materials for energy storage; and simulation of multi-phase fluids, and combustion. To ensure the impact of this work on the scientific community, we will disseminate these codes on github and the educational materials we produce on a dedicated website.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210107
Entities
People
- Sigal Gottlieb
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Massachusetts