GPUnion: Autonomous GPU Sharing on Campus

2025-07-24

AI Summary

To address uneven distribution and low utilization of GPU resources across campus laboratories, this paper designs and implements a decentralized GPU resource-sharing platform. The platform adopts a “provider-first” architecture to ensure laboratories retain full autonomy over their owned GPUs. It enables cross-laboratory resource collaboration under strong isolation and high security via non-root containerized scheduling, image attestation, custom storage integration, and elastic checkpoint-based migration. A key innovation is automated task migration upon provider disconnection—achieving a 94% success rate and significantly enhancing system robustness. Experimental evaluation demonstrates a 30% increase in average GPU utilization and a 40% rise in interactive sessions, thereby improving accessibility and sustainability of AI research infrastructure.

Technology Category

• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Machine Learning: Hardware-aware ML
• Computer Vision: Remote Sensing / Geospatial AI

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Decentralized Web and Fediverse systems
• Economics, Online Markets and Human Computation: The sharing economy
• Security and Privacy: Data transparency and provenance

Abstract

A pronounced imbalance in GPU resources exists on campus, where some laboratories own underutilized servers while others lack the compute needed for AI research. GPU sharing can alleviate this disparity, while existing platforms typically rely on centralized oversight and persistent allocation models, conflicting with the voluntary and autonomous nature of academic resource ownership. We present GPUnion, a campus-scale GPU sharing platform enabling voluntary participation while preserving full provider autonomy. GPUnion incorporates three core mechanisms: i) container-based task dispatching and execution, ii) resource provider-first architecture, and iii) resilient execution featuring automatic checkpointing and migration. GPUnion also supports custom data storage and integrates the non-root execution and image attestation for isolation and security improvement for containerization. Case studies across multiple campus scenarios demonstrate 30% more GPU utilization improvement, 40% increase in interactive sessions, and 94% successful workload migration during provider departures. GPUnion demonstrates that provider autonomy and platform reliability can coexist, challenging conventional centralized paradigms and democratizing access to computational resources within campus networks.

Problem

• Addresses GPU resource imbalance on campus for AI research
• Enables autonomous GPU sharing without centralized oversight
• Improves GPU utilization and workload migration reliability

Innovation

• Container-based task dispatching and execution
• Resource provider-first architecture design
• Automatic checkpointing and migration resilience