Hybrid Learning and Optimization-Based Dynamic Scheduling for DL Workloads on Heterogeneous GPU Clusters

2025-12-10

AI Summary

Scheduling deep learning jobs on heterogeneous GPU clusters faces challenges including lack of job profiling, application-agnostic requirements, and dynamic workload variations. Method: This paper proposes RLTune—the first offline-profilings-free, application-agnostic real-time scheduling framework. It jointly optimizes job completion time, queueing delay, and GPU utilization via end-to-end integration of Proximal Policy Optimization (PPO) for online priority assignment and Mixed-Integer Linear Programming (MILP) for precise node mapping. Contribution/Results: RLTune establishes the first system-level co-design paradigm integrating RL policies with MILP for cluster scheduling. It generalizes zero-shot across production traces from diverse platforms (Philly, Helios, Alibaba). Experiments demonstrate up to 20% higher GPU utilization, 81% lower queueing delay, and 70% reduced job completion time. Deployment feasibility is further validated in multi-cloud environments.

Technology Category

• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Modern cloud platforms increasingly host large-scale deep learning (DL) workloads, demanding high-throughput, low-latency GPU scheduling. However, the growing heterogeneity of GPU clusters and limited visibility into application characteristics pose major challenges for existing schedulers, which often rely on offline profiling or application-specific assumptions. We present RLTune, an application-agnostic reinforcement learning (RL)-based scheduling framework that dynamically prioritizes and allocates DL jobs on heterogeneous GPU clusters. RLTune integrates RL-driven prioritization with MILP-based job-to-node mapping to optimize system-wide objectives such as job completion time (JCT), queueing delay, and resource utilization. Trained on large-scale production traces from Microsoft Philly, Helios, and Alibaba, RLTune improves GPU utilization by up to 20%, reduces queueing delay by up to 81%, and shortens JCT by as much as 70 percent. Unlike prior approaches, RLTune generalizes across diverse workloads without requiring per-job profiling, making it practical for cloud providers to deploy at scale for more efficient, fair, and sustainable DL workload management.

Problem

• Optimizes scheduling for DL workloads on heterogeneous GPU clusters
• Reduces job completion time and queueing delay without per-job profiling
• Improves GPU utilization and generalizes across diverse application workloads

Innovation

• Reinforcement learning-based dynamic scheduling for heterogeneous GPU clusters
• Integrates RL prioritization with MILP-based job-to-node mapping
• Generalizes across workloads without per-job profiling for scalable deployment