Optimization Study for Multicores

Abstract: IT giants like Intel and AMD have set the stage for extensive use of Multicoreprocessors in IT business and research, thereby providing much more computingpower to run software for both the scientific and commercial use. One of the problems with writing software in traditional fashion is that Softwareengineers and even top architects may not take into consideration architecture-basedor resource-aware optimizations, simply relying on plain process based parallelism toimprove performance of the target application for Multi-Core processors. Trueoptimal performance for any given parallelized application can be achieved only whenit has been carefully written to utilize the available resources with a deeperunderstanding of the underlying available hardware resources and processorarchitecture e.g. memory bandwidth, cache architecture etc. The purpose of this study is to demonstrate performance optimization techniques forapplications from various fields, parallelization techniques for these applications andthe maximal performance achieved for parallelized applications over multicoreprocessors through architecture and resource aware optimizations. We show howthe simplest of code optimizations help improve application performance multipletimes, speeding up the applications by a factor of more than two.