.png){kind=icon}
.png){kind=icon}
Return
Dynamic Multi-task Flow Scheduling Model Based on Heterogeneous Computing Platform
.png){kind=icon}
PRE
AI
DOI：10.1007/s13369-026-11474-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Heterogeneous computing platforms provide diverse processing units for large-scale workloads, yet efficient scheduling remains challenging due to cross-platform performance disparity, dynamic resource contention, and the lack of fine-grained task characterization. This paper presents the HETEROPERF framework to capture runtime performance characteristics and heterogeneous resource requirements, enabling accurate task–resource matching. Based on these representations, a deep reinforcement learning (DRL)-based scheduling approach is devised using Soft Actor–Critic (SAC) for dynamic multi-task flows, and learning efficiency is further improved via prioritized experience replay. Extensive simulation results demonstrate that the proposed method consistently achieves superior scheduling performance compared with representative baselines across diverse scheduling scenarios.
Keywords:
Heterogeneous computing platforms
Task characterization
Task flow scheduling
DRL
Journal
A
IF:
2.9
Papers:
962
Citations:
0