Guardian: Safe GPU Sharing in Multi-Tenant Environments
Modern GPU applications, such as machine learning (ML), can only partially utilize GPUs, leading to GPU underutilization in cloud environments. Sharing GPUs across multiple applications from different tenants can improve resource utilization and consequently cost, energy, and power efficiency. However, GPU sharing creates memory safety concerns because kernels must share a single GPU address space. Existing spatial-sharing mechanisms either lack fault isolation for memory accesses or require static partitioning, which leads to limited deployability or low utilization.
Authors
- Giorgos Vasiliadis (ORCID: https://orcid.org/0000-0001-5470-4714)
- Stelios Mavridis (ORCID: https://orcid.org/0000-0001-9857-9185)
- Manos Pavlidakis (ORCID: https://orcid.org/0000-0002-8341-3083)
- Angelos Bilas (ORCID: https://orcid.org/0000-0003-2975-4124)
- Antony Chazapis (ORCID: https://orcid.org/0000-0002-4729-7396)
- Anargyros Argyros
Institutions
- University of Crete (GR)
- Mediterranean University (ME)
- Hellenic Mediterranean University (GR)
- FORTH Institute of Computer Science (GR)
- Foundation for Research and Technology Hellas (GR)
Publication Details
- Published
- 2024-11-27
- DOI
- https://doi.org/10.1145/3652892.3700768
- Citations
- 7
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 1.74