Skip to Content

Parallel-ARD Solver for the Acoustic Wave Equation

Dinesh Manocha, University of North Carolina, Chapel Hill

Usage Details

Blue Waters Trainee 208, Blue Waters Trainee 209, Blue Waters Instructor 003, Dinesh Manocha, Vivek Chavda

We plan to test the performance of our Parallel-ARD simulator on many distributed processors and measure the speedup gained by parallelizing over more cores. Our acoustic simulator is based on a novel method to solve the acoustic wave equation. In particular, we compute an adaptive rectangular decomposition of the domain and perform the computations in parallel across a large number of cores. Small variations in air pressure—the source of sound—are governed by the 3D wave equation, a second-order linear partial differential equation. The computational complexity of solving this wave equation increases as at least the cube of frequency, and it is a linear function of the volume of the scene. Given the aural range of humans (20Hz–22kHz), performing acoustic simulation for acoustic spaces corresponding to a concert hall or a cathedral (e.g., volume of 10,000–15,000 m3) for the maximum simulation frequency of 22kHz requires tens of exaflops of computational power and tens of terabytes of memory.

We will use the Blue Waters supercomputer system to help measure the scalability of our algorithm on a high number of cores. This project will provide necessary exposure and experience to an undergraduate Blue Waters intern, Vivek Chavda, in terms of writing and evaluating parallel programs for scientific computation. By testing the performance on the Blue Waters system, we will have sufficient reliable data to measure the parallel performance of our algorithm and improve its performance by making appropriate changes to the algorithm and the code.