high performance computing on graphics processing units: hgpu.org

The environment around us has many phenomena and has different behaviors according to different parameters, biological, chemical, physical, etc. To represent a simple and abstract reality of this environment we use a concept called environmental modeling. The environmental modeling deals with many environmental problems such as air pollution, diffusion of disease, animal behavior and so on. However there are some difficulties in modeling the environment due to several reasons such as the variation and incompatibility. There are many techniques to simulate a model such as differential equations, cellular automata and multi-agent approach. We will focus on our work on the cellular automata, that is dynamic model which is spatially and temporally spate. The cellular automate update their state with time, where each cell has a new state based on its state and the state of its neighbors. The update is based also on some rules defined previously. When simulating any model, the most important problem that faces us is the execution time. In many environmental problems we will work on a model that may cover a big geographical area and thus we require a large numbers of cells to represent that system. For these cells to update their state we will need a long execution time which may exceed days in some models. According to Florent Arrignon, a specialist in environmental modeling, in some models of one million cells, it requires about one week to simulate it. But fortunately due to the increased performance in the computers and more precisely on the GPU, we can simulate a cellular automata environmental model in parallel which will result in enhanced performance result. Besides, there are different approaches to deal with large blocks of data such as OpenCL and CUDA. In our case study we are addressing a general problem of Forest Fire Spread using CUDA language and discuss different approaches to enhance the performance.