Enabling easey deployment of containerized applications for future hpc systems

Abstract

The upcoming exascale era will push the changes in computing architecture from classical CPU-based systems towards hybrid GPU-heavy systems with much higher levels of complexity. While such clusters are expected to improve the performance of certain optimized HPC applications, it will also increase the difficulties for those users who have yet to adapt their codes or are starting from scratch with new programming paradigms. Since there are still no comprehensive automatic assistance mechanisms to enhance application performance on such systems, we propose a support framework for future HPC architectures, called EASEY (Enable exASclae for EverYone). Our solution builds on a layered software architecture, which offers different mechanisms on each layer for different tasks of tuning, including a workflow management system. This enables users to adjust the parameters on each of the layers, thereby enhancing specific characteristics of their codes. We introduce the framework with a Charliecloud-based solution, showcasing the LULESH benchmark on the upper layers of our framework. Our approach can automatically deploy optimized container computations with negligible overhead and at the same time reduce the time a scientist needs to spent on manual job submission configurations.