Modeling Parallel Applications for Scalability Analysis: An approach to predict the communication pattern

摘要

The performance of message-passing applications varies depending on the parallel system, causing potential inefficiencies when its number of processes increases. By this reason, it is critical to predict the application behavior before executing it, in order to use the system efficiently. We propose a methodology that allows us to predict the application scalability behavior in a specific system, providing information to select the most appropriate resources to run the application. The methodology strives to use a bounded analysis time, and a reduced set of resources. This paper presents the general methodology, focusing on validating the step of the methodology concerning to the generation of scalability communication model. We can predict the evolution of the communication pattern using a reduced set of resources. Analyzing from 16 to 256 processes, we can predict the communication pattern for 4,096 processes.