GPU parallelization of brownout simulations with a non-interacting particles dynamic model
Keywords
High Performance Computing
SC4I/Digitization, Innovation, and Competitiveness of the Production System
Code:
29/2016
Title:
GPU parallelization of brownout simulations with a non-interacting particles dynamic model
Date:
Saturday 30th July 2016
Author(s):
Miglio, E.; Parolini, N.; Penati, M.; Porcù, R.
Download link:
Abstract:
The term brownout refers to the uplift of sand particles in the air and is
generated when a helicopter is close to a dusty soil. When a brownout occurs the visibility area is remarkably restricted, thus the pilot may be disoriented and the helicopter may dangerously collide with the ground. Simulations of a brownout require tens of millions of particles in order to be significative, so that the execution of a serial program takes a very long time. In order to speedup the computation, the GPU-parallelization of a brownout simulation program is performed in order to obtain a notable speedup. The dynamics of the particles are considered in a Lagrangian way, under the effect of the gravity force and of a precomputed aerodynamic field. The particles are independent from each other since collisions between them are not taken into account. Thus trajectories are independent and the parallelization is very effective. In this paper we discuss in detail the impact of the techniques used for the GPU implementation of the parallel code on the performance.