Abordagem para reduzir tempo de execução de aplicações paralelas para análise de escalabilidade

Abstract

The increase in computational power and the use of parallelism techniques are fundamental strategies for solving increasingly complex problems. However, to use the resources of these systems efficiently, it is essential to perform performance analysis. Thus, performance analysis tools are essential for assisting in optimization and identifying bottlenecks. Additionally, scalability analysis is crucial in this context, as it allows for evaluating how the application behaves as computational resources or workload increase. In this way, evaluating scalability is useful for predicting how the application will behave in larger systems or situations of increased load. It is worth noting that these analyses may require execution time that can scale with the complexity of the application. In this context, the objective of this work is to implement a partial execution approach to estimate the total execution time of applications without the need to execute them completely, focusing only on a portion of the iterations of the main loop, in the Pascal Analyzer scalability analysis tool. The implementation was carried out in the C++ programming language, using strategies to maintain data consistency. The results obtained from the tests showed that the developed function was efficient in reducing execution time in various configurations and different portions of iterations, presenting minimal errors in most cases, although it presented high precision errors in some configurations. Overall, for the purposes of this work, the implementation and results were satisfactory.