On limits of Machine Learning techniques in the learning of scheduling policies

• Grant number: 22/06906-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: July 01, 2022
• End date: July 31, 2023
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Alfredo Goldman vel Lejbman
• Grantee: Lucas de Sousa Rosa
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 19/26702-8 - Trends on high performance computing, from resource management to new computer architectures, AP.TEM
• Associated scholarship(s): 22/14673-6 - Evaluating machine learning techniques and simulations to create efficient scheduling heuristics, BE.EP.IC

Abstract

The use of High Performance Computing (HPC) has been increasing in the most different areas. At the same time that such platforms evolve in both size and complexity; its energy consumption also increases. In this sense, efforts are being made by researchers in an attempt to make a more efficient use of these platforms. In this work, we propose to contribute to these efforts by improving one of the different aspects related to performance: the resource management.We intend to explore different methodologies using Machine Learning in the context of job scheduling. Starting from data generated by simulation, we intend to model the impact of scheduling different jobs in terms of the average bounded slowdown through regression methods. Finally, each model will be tested in simulated experiments considering different platform configurations and jobs characteristics.