Dynamic load balance applied to a numerical weather forecast model accelerated by using GPUs

• Grant number: 12/12144-4
• Support Opportunities: Scholarships abroad - Research
• Start date: January 07, 2013
• End date: July 06, 2013
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Álvaro Luiz Fazenda
• Grantee: Álvaro Luiz Fazenda
• Host Investigator: Laxmikant V. Kale
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil
• Institution abroad: University of Illinois at Urbana-Champaign, United States

Abstract

This research project aims to investigate dynamic load balance by using process virtualization / migration applied to a scientific application that would be ported to use GPGPU ("General-Purpose computation on Graphics Processing Units") as performance accelerator. The application chosen is an open-source, limited-area numerical weather forecast model: BRAMS ("Brazilian developments on the Regional Atmospheric Modeling System"), developed and supported by CPTEC/INPE (Center for Weather Forecast and Climatic Studies/National Institute for Space Research). The plans include porting part of the application with significant computational load demands to use GPGPUs by employing either the CUDA language or the OpenACC standard, and also investigate dynamic load balancing by using AMPI/Charm++, developed at UIUC (University of Illinois at Urbana-Champaign) and already tested with BRAMS in homogeneous systems. The main activities would occur at UIUC, where there is access to modern supercomputing facilities that are appropriate for the planned tasks. (AU)