A Preliminary Study on Reinforcement Learning and Genetic Algorithm for Job Scheduling

This paper presents a comparative study between Reinforcement Learning (RL) and
Genetic Algorithm (GA) for the job scheduling problem in supercomputers. We evaluate the per-
formance of both techniques using logs from three real supercomputers, focusing on the bounded
slowdown metric. The results demonstrate that, for the implemented algorithms, the GA stands
out for its low computational cost, obtaining high-quality solutions in a timely manner, while RL
achieves superior performance, but at a significantly higher computational cost.