Deep Reinforcement Learning Tested for HPC Job Scheduling

A new research paper evaluates the effectiveness of six Deep Reinforcement Learning (DRL) treatments for High-Performance Computing (HPC) job scheduling. The study compared Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), and Deep Q-Network (DQN), each with and without action masking, against traditional heuristics and a masked random control. Results indicate that while masked DRL treatments generally completed evaluations, only MaskablePPO showed significant improvements in average waiting time and maximum bounded slowdown on a heterogeneous workload trace compared to the control. AI

IMPACT This research provides insights into the practical application of DRL for optimizing HPC resource allocation, potentially improving efficiency in large-scale computing environments.

RANK_REASON Academic paper detailing a statistical evaluation of DRL methods for HPC job scheduling. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Advantage Actor-Critic
• arXiv
• Deep Q-Network
• deep reinforcement learning
• high-performance computing
• Hugging Face
• Last-Come, First-Served
• MaskableA2C
• MaskableDQN
• MaskablePPO
• Proximal Policy Optimization
• Slurm

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →