Exploring the Epiphany manycore architecturefor the Lattice Boltzmann algorithm

Computational fluid dynamics (CFD) plays an important role in many scientific applications, ranging from designing more effective boat engines or aircraft wings to predicting tomorrow's weather, but at the cost of requiring huge amounts of computing time. Also, traditional algorithms suffer from scalability limitations, making them hard to parallelize massively. As a relatively new and promising method for computational fluid dynamics, the Lattice Boltzmann algorithm tries to solve the scalability problems of conventional, but well-tested algorithms in computational fluid dynamics. Through its inherently local structure, it is well suited for parallel processing, and has been implemented on many different kinds of parallel platforms. Adapteva's Epiphany platform is a modern, low-power manycore architecture, which is designed to scale up to thousands of cores, and has even more ambitious plans for the future. Hardware support for floating-point calculations makes it a possible choice in scientific settings. The goal of this thesis is to analyze the performance of the Lattice Boltzmann algorithm on the Epiphany platform. This is done by implementing and testing the lid cavity test case in two and three dimensions. In real applications, high performance on large lattices with millions of nodes is very important. Although the tested Epiphany implementation scales very good, the hardware does not provide adequate amounts of local memory and external memory bandwidth, currently preventing widespread use in computational fluid dynamics