Improving Analysis in SPMD Applications for Performance Prediction

The analysis of parallel scientific applications allows us to know the details of their behavior. One way of obtaining information is through performance tools. One such tool is PAS2P, which is based on parallel application repeatability, focusing on performance analysis and prediction using the application signature. The analysis is performed using the same execution resources of the parallel application to create an independent machine model and identify common patterns. The analysis stage of the PAS2P tool is costly in terms of runtime, due to the high number of communications it performs, degrading performance by increasing the number of execution processes. To solve this problem, we propose designing a module that reduces the data dependency between processes, reducing the number of communications, and taking advantage of the characteristics of the SPMD applications. For this, we propose an analyzer module that is independent of data between processes. Our proposal allowed us to decrease the analysis time when the application scales.