Performance analisys on multicore system using PAPI

摘要

The rapid growth of multicore systems, and approaches that they have taken, allow complex processes that before were only possible to run on supercomputers, can now be run on low-cost solutions also called "commodity hardware". Such solutions can be implemented using processors consumer market (Intel and AMD). When these solutions scale to scientific computing requirements, is essential to have methods to measure the performance that they offer and how these behave under different workloads. Due to the large number of load types on the market, and even within the scientific computing, it is necessary to introduce "typical workload" that can serve as support in the acquisition and evaluation processes solutions, having a high degree of certainty on this operation. In this research proposes a practical approach to the evaluation and presents the results of tests performed on equipment AMD and Intel multicore architectures