LGAIDCMay 26

When Does Deep RL Beat Calibrated Baselines? A Benchmark Study on Adaptive Resource Control

arXiv:2605.2641840.0
AI Analysis

For practitioners in adaptive resource control, this paper shows that simple baselines often beat DRL, shifting focus from algorithm selection to calibration and evaluation.

A calibrated rule-based autoscaler outperforms six deep RL algorithms on cost across all tested workloads, though RL agents show promise on bursty traffic. The study finds that discrete-action algorithms have far fewer constraint violations than continuous-action ones, and no single RL algorithm dominates across workloads.

A properly calibrated rule-based autoscaler can beat every one of six mainstream deep reinforcement learning (DRL) algorithms on cost across every workload we test - so when, if ever, does DRL actually help? We study this in RLScale-Bench, a reproducible benchmark and evaluation protocol for DRL on adaptive resource control, where an agent allocates compute to a dynamic workload under cost and service-level constraints. We evaluate PPO, DQN, A2C, SAC, TD3, and DDPG under matched architectures, training budgets, and reward functions against a calibrated rule-based baseline across six workload patterns and five seeds (240 runs), instantiate the benchmark on Kubernetes Horizontal Pod Autoscaling, and probe distribution-shift generalization. Three findings challenge common assumptions: (i) the calibrated controller achieves the lowest cost on all six workloads, though it trails the best RL agents on bursty and flash traffic; (ii) discrete-action algorithms outperform continuous-action ones by one to two orders of magnitude in constraint violations due to action-space mismatch; and (iii) no single algorithm dominates across workloads, with rankings shifting by up to four positions. The bottleneck in RL-based resource control is not algorithm selection but baseline calibration, reward engineering, and realistic evaluation protocols.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes