.png){kind=icon}
.png){kind=icon}
Return
STAO: Adaptive spatio-temporal orchestration for efficient and QoS-aware DNN inference on GPU
.png){kind=icon}
PRE
AI
S
X
K
J
T
DOI：10.1016/j.future.2026.108734.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Introduces transparent spatial partitioning mechanism enabling dynamic, fine-grained GPU resource adjustment without application modification. • Proposes an interference-aware latency model to accurately assess QoS risks and guide prediction-driven online scheduling decisions. • Implements kernel-level command buffer interception to achieve robust temporal multiplexing and strict isolation across GPU generations. • Develops STAO, a spatio-temporal system that significantly improves throughput and eliminates QoS violations in DNN co-location.
Journal
F
IF:
6.1
Papers:
6.8K
Citations:
2.3W
Organization
No organization information available
Cited Papers