Accelerate SOMA using parallel processing in GPGPU

Tác giả: Dao T.T., Toan N.M., Duy V.H., Zelinka I.

Abstract: This paper presents methods for implementing SOMA (Self-Organizing Migrating Algorithm) in parallel with the CUDA (Compute Unified Device Architecture) system that can be used to perform the dominant of up-speed when using SOMA algorithm. SOMA has many individual points to find the global minimum which is the key for paralleling this system because each individual can work separately and share the position for all when it moves. Nowadays, due to the humongous size of data and the limitation of the process in single Central Processing Unit (CPU), it becomes impossible to deal with. As a result of these limitations, we need more CPUs working at the same time to do the same job or take advantage of the power of parallel processing in GPGPU (General-Purpose graphics processing unit). Additionally, many supercomputers are built with the need of Parallel Processing in order to meet the power of hardware. Based on the architecture of CUDA, it can handle the threads in SOMA independence. We use two methods with different architecture in CUDA to help SOMA run much faster than single threading method. This paper also uses some techniques to help SOMA work more effective.