Advances, Systems and Applications
Table 7 Scheduling techniques for traditional and AI workloads
From: Fast DRL-based scheduler configuration tuning for reducing tail latency in edge-cloud jobs
Category | Introduction |
|---|---|
Cluster resource management systems | Provide scheduling configuration parameters (e.g. Google Borg [66],YARN [65], Mesos [32], Kubernetes [19]) |
Scheduling optimization for traditional workloads | Apply DRL under a simplified state space (e.g. DeepRM [44]) |
| Â | Adopt event-driven decision framework to reduce the booming action space (e.g. [9, 13, 41, 70]) |
| Â | Apply DRL to tackle DAG-based job scheduling problems (e.g. Decima [45], Spear [34]) |
Scheduling optimization for AI workloads | Take the feature of iterativeness into consideration (e.g. SLAQ [79], Optium [51], OASiS [6]) |
| Â | Adopt DRL techniques into job scheduling optimization (e.g. Harmony [5], SIREN [67], DSS [57]) |
| Â | Apply DRL to optimize the task scheduling (e.g. [15, 38, 72]) |