The joint paper “An Exploratory Study of Deep Learning for Predicting Computational Tasks Behavior in HPC Systems” (https://www.computer.org/csdl/proceedings-article/sbac-padw/2023/816000a009/1RVZt2fyTjG) was presented in the HPC/DL workshop during the SBAC-PAD conference in October 2023, in Porto Alegre, Brazil.
The paper explores the use of machine learning techniques (including regression and neural networks) to predict usage metrics (such as execution time and memory usage) of jobs in the Santos Dumont machine. This research is part of an ongoing effort to automatically tune submission parameters of jobs coming from the BioInfoPortal.
The BioInfoPortal offers BioInformatics applications and workflows, which users can ask to execute through the web interface. These executions are submitted as jobs to the Santos Dumont system. The goal is to eventually be able to automatically set the configuration (number of nodes, of processes, etc) for each job coming from the BioInfoPortal aiming at providing good performance and, more importantly, efficiently using the HPC resources.
