Halo Gathering Scalability for Large Scale Multi-dimensional Sznajd Opinion Models Using Data Parallelism with GPUs

摘要

The Sznajd model of opinion formation exhibits complex phase transitional and growth behaviour and can be studied with numerical simulations on a number of different network structures. Large system sizes and detailed statistical sampling of the model both require data-parallel computing to accelerate simulation performance. Data structures and computational performance issues are reported for simulations on single and multi-core processing devices. A discussion of optimal data structures for performance on Graphical Processing Units using NVIDIA'S Compute Unified Device Architecture (CUDA) is also given. System size and memory layout tradeoffs for different processing devices are also presented.