Optimization of K means clustering using parallel programming on graphics processing unit

Abstract

High Performance Computing has evolved as a tool to meet the growing computational requirements of existing and emerging scientific and engineering applications. With advancements in technology that enable the acquisition and storage of large amounts of data, several of these applications are becoming data-driven and are involved in the access and analysis of huge data sets. To meet the rising performance demands of these applications on modern high performance computing systems, it is crucial to design algorithms for their efficient execution. Data accumulates in immense quantity in the real time and parallel data mining techniques are required to analyze massive databases in a reasonable time. Clustering high performance data analysis applications involves several challenges. These applications are characterized by significant computation, communication as well as Input / Output requirements. The algorithms in data clustering are costly with high demands for computation and data access, which can be handled efficiently by parallel computers. Since designing and implementing parallel programs on parallel computers is expensive an alternative was needed which lead way to General Purpose Graphics Processing Units. The Graphics Processing Unit is a device to accelerate particular program code so that programs start execution on the Central Processing Unit and launch procedures of code referred to as kernels onto the Graphics Processing Unit to get accelerated. The Compute Unified Device Architecture is a framework for scientific general purpose computing on NVIDIA (Nvidia Corporation) Graphics Processing Units. newline