Research map / Reinforcement learning / Visual and representation RL
Scheduling of Time-Varying Workloads Using Reinforcement Learning
Shanka Subhra Mondal, Nikhil Sheoran, Subrata Mitra
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use temporal resource histories and reinforcement learning for placement | TVW-RL · Google and Alibaba traces | average utilization vs. DeepRM: 30-300% | Improved |
| use temporal resource histories and reinforcement learning for placement | TVW-RL · Google and Alibaba traces | average utilization vs. Tetris: 68-100% | Improved |
| use temporal resource histories and reinforcement learning for placement | TVW-RL · Google and Alibaba traces | resource fragmentation vs. DeepRM: 5-50% | Improved |
| use temporal resource histories and reinforcement learning for placement | TVW-RL · Google and Alibaba traces | resource fragmentation vs. Tetris: 6-13% | Improved |
| use temporal features for workload equivalence classes instead of aggregate features (ablation) | TVW-RL workload mapping · Google and Alibaba traces, usage prediction | effective prediction gain: 6.82 -> 9.83; 5.13 -> 8.14 | Improved |
| remove equivalence-class mapping (ablation) | TVW-RL · Google traces, 50% load | CPU utilization: 0.328 vs. 0.345 | Hurt |
| increase the under-utilization penalty Ku (ablation) | TVW-RL · Alibaba traces, 50% load | CPU utilization: 0.103 -> 0.179 | Improved |
| increase the overshoot penalty Ko (ablation) | TVW-RL · Alibaba traces, 50% load | resource overshoot: 2.20, 0.20, 0.15, 0.09% | Improved |
| increase the resource contention penalty Kc (ablation) | TVW-RL · Alibaba traces, 50% load | CPU and memory utilization: 0.17 to 0.12; 0.28 to 0.2 | Hurt |
| train with noisy background workloads rather than without them (ablation) | TVW-RL · Google traces, 50% load, noise [0-3] | percentage overshoot: 0.055 vs. 0.132 | Improved |
| add discrete uniform noise in [-1,1] to workload shapes (ablation) | TVW-RL · Google traces, noisy workload shapes | CPU and memory utilization: 2%; 7% | Hurt |
Closest papers by meaning
- Inductive-bias-driven Reinforcement Learning For Efficient Schedules in Heterogeneous Clusters (ICML 2020 · similarity 0.87)
- Deft Scheduling of Dynamic Cloud Workflows with Varying Deadlines via Mixture-of-Experts (ICLR 2026 · similarity 0.86)
- Deadline-Constrained Dynamic Workflow Scheduling Can be Cast as a Representation Learning Problem (NeurIPS 2026 · similarity 0.86)
- DCcluster-Opt: Benchmarking Dynamic Multi-Objective Optimization for Geo-Distributed Data Center Workloads (NeurIPS 2025 · similarity 0.84)
- Learning Memory-Enhanced Improvement Heuristics for Flexible Job Shop Scheduling (NeurIPS 2025 · similarity 0.83)
- SustainDC: Benchmarking for Sustainable Data Center Control (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
