Home // International Journal On Advances in Systems and Measurements, volume 13, numbers 3 and 4, 2020 // View article

An Agent-Based Model for Analyzing the HPC Input/Output System

Authors:
Diego Encinas
Sandra Mendez
Marcelo Naiouf
Armando De Giusti
Dolores Rexachs
Emilio Luque

Keywords: Agent-Based Modelling and Simulation (ABMS); HPC-I/O System; Parallel File System.

Abstract:
High Performance Computing (HPC) applications can spend a significant portion of their execution time doing Input/Output (I/O) operations into files. Improving I/O performance becomes more important for the HPC community, as parallel applications produce more data and use more computing resources. One of the methods used to evaluate and understand the I/O performance behavior of such applications in new I/O systems or for different configurations is using modeling and simulation techniques. In this paper, we present a simulation model of the HPC I/O system by using Agent-Based Modeling and Simulation (ABMS) based on the functionality of the I/O Software Stack. Our proposal is modeled using the concept of white box so that the specific behavior of each of the modules or layers in the system can be observed. The interaction between the layers of the I/O software stack are analyzed by monitoring the internal functions using proprietary parallel file system tools. This allows obtaining the functional and temporal characteristics corresponding to the I/O operations. These characteristics allowed the design and implementation of a representative model of I/O system components. Furthermore, measurements are used to obtain the necessary data sets in the verification, fine-tuning and validation stages. The resulting implementation has shown similar behaviors for measured and simulated values when using the IOR benchmark with various file sizes.

Pages: 192 to 202

Copyright: Copyright (c) to authors, 2020. Used with permission.

Publication date: December 30, 2020

Published in: journal

ISSN: 1942-261x