Peta-Flops Acoustic Simulation
Principal Investigator:Dinesh Manocha
Funding Agency: National Science Foundation
Agency Number:OCI-0904990
Abstract
The objective of this proposal is to develop new petascale algorithms for acoustic simulation. The main focus will be on developing massive parallel algorithms that can exploit the computational capabilities of many-core accelerators such as graphics processing units (GPUs) to achieve the petaflop performance goals. The proposed research includes development of accurate numerical methods for the wave equation and fast solutions based on sound field decomposition. Moreover, the PIs will develop parallel techniques to perform these computations using O(100, 000) threads concurrently on multiple many-core accelerators. The interdisciplinary team of PIs brings a strong background in scientific computing, computational acoustics, parallel algorithms, and high performance computing using GPUs to address this challenging problem. The PIs will work closely with leading industrial players to develop petaflop computation capabilities and test the performance of the resulting simulation on CAD models of airplane cabins.
INTELLECTUAL MERIT:
This research is expected to make two major advancements. Firstly, it could enable performing accurate acoustic simulation on large models and handle high-frequency acoustic sources. Secondly, it could result in novel architectures for petascale computing that exploit the many-core capabilities of commodity accelerators such as GPUs. The specific contributions could include: (1) Solving high-frequency acoustic waves using highly accurate and low dispersion numerical methods; (2) Acoustic simulation of low and medium frequency ranges using adaptive rectangular decomposition; (3) Massively parallel algorithms and software libraries for 3D FFT and numerical linear algebra; (4) Parallel algorithms for spatial decomposition of sound field decomposition and geometric propagation; (5) Efficient software libraries to run hundred of thousands of threads to achieve petaflop performance; (6) Accurate acoustic simulation of complex CAD models.
In addition to computational acoustics, the proposed research could also offer fundamental advances and petascale capability for scientific problems in geophysics, meteorology, electromagnetics, etc.
BROADER IMPACT: The proposed research could make a significant impact on the following:
Acoustic Simulation of Complex CAD models: The acoustic software simulation market is currently estimated at $500M a year and is used for virtual prototyping airplanes, automobiles, auditoriums and urban environments. The petaflop algorithms and simulators could significantly push the state of the art in terms of handling complex models and could potentially result in savings of millions of dollars.
Accurate solvers for high-frequency wave equation: This is regarded as one of the most challenging unsolved problems in scientific computing and finite element methods. Our solution in terms of highly accurate and low dispersion numerical methods could offer new scientific insights. Furthermore, our formulation based on sound field decomposition and geometric propagation could provide a good approximate solution on large, complex models.
Massively parallel computing using many-core accelerators: The computing industry is undergoing a major paradigm shift as current and future processors consist of multiple cores. As a result, there is a significant need for developing new parallel algorithms and libraries to exploit the hardware capabilities. The proposed research could potentially demonstrate the benefit of using many-core accelerators for many scientific applications and achieve petascale performance by running a very large number of threads concurrently.
Software libraries: As an integral part of this research, we will develop and distribute new many-core libraries for numerical algorithms, including linear algebra solvers and 3D FFT, and geometric propagation using ray tracing and hierarchical data structures.
Education and Outreach: We will develop two new courses, including an undergraduate course on parallel scientific computation and a graduate inter-disciplinary course on acoustic simulation. We will organize workshops and focused sessions on these technologies at the leading conferences on high performance computing, scientific computing and computational acoustics. We will also expose these technologies to middle and high school students by developing some creative applications using our acoustic simulator