Experimental Dataset for Control-Aware Autoscaling of Shared Controllers in Edge Kubernetes Systems

This dataset provides a comprehensive experimental benchmark for the study of control-aware autoscaling of shared controllers in edge computing environments based on Kubernetes. It contains aggregated measurements from a large set of controlled experiments in which multiple Industrial Internet of Things (IIoT) devices share a limited number of fuzzy control services deployed as scalable Kubernetes pods. The experiments systematically vary the number of devices, the number of shared controllers, and the operating regime of the controlled systems, enabling the analysis of trade-offs between control performance, temporal stability, and computational resource usage. The dataset covers three representative classes of control systems with distinct dynamics: a DC motor (fast and delay-sensitive dynamics), a level control tank (slow and naturally damped dynamics), and a vehicle speed control system (high-inertia dynamics). For each plant, multiple operating setpoints were evaluated, and for each setpoint the number of active IIoT devices and shared controllers was varied across a wide range. The dataset includes key metrics such as the integrated absolute error (IAE), the 90th percentile of control loop time and request delay, CPU usage of the control services, and network traffic statistics. Each row in the dataset corresponds to a complete experimental scenario, fully characterized by its control, computational, and networking context. This dataset is intended to support reproducible research in edge control systems, autoscaling strategies, and data-driven resource management. It can be used for empirical performance evaluation, comparative analysis of autoscaling policies, and as a benchmark for machine learning methods such as risk-aware regression, quantile prediction, and offline reinforcement learning. By providing a unified and well-documented experimental dataset, this work aims to facilitate further research on scalable control architectures and intelligent resource management in edge computing environments.本数据集为基于Kubernetes的边缘计算环境中共享控制器的控制感知自动扩缩容研究提供了一套全面的实验基准测试集。数据集包含大量受控实验的聚合测量数据，这些实验中，多台工业物联网（Industrial Internet of Things，IIoT）设备共享以可扩展Kubernetes Pod形式部署的有限数量模糊控制服务。本实验集通过系统性改变设备数量、共享控制器数量以及被控系统的运行模式，支持对控制性能、时间稳定性与计算资源使用率之间权衡关系的分析。 本数据集涵盖三类具有不同动态特性的典型被控系统：直流电机（DC motor，快速且对延迟敏感的动态特性）、液位控制水箱（慢速且自然阻尼的动态特性）以及车辆速度控制系统（高惯性动态特性）。针对每一类被控对象，我们评估了多种运行设定点；针对每个设定点，活跃IIoT设备数量与共享控制器数量均在较大范围内变化。数据集包含多项关键指标，如积分绝对误差（IAE）、控制回路时间与请求延迟的90分位数、控制服务的CPU使用率以及网络流量统计数据。数据集中的每一行对应一个完整的实验场景，其特征由控制、计算与网络上下文全面刻画。 本数据集旨在支持边缘控制系统、自动扩缩容策略以及数据驱动资源管理领域的可复现研究。其可用于实证性能评估、自动扩缩容策略的对比分析，还可作为风险感知回归、分位数预测、离线强化学习等机器学习方法的基准测试集。本工作通过提供一套统一且文档完善的实验数据集，旨在推动边缘计算环境下可扩展控制架构与智能资源管理领域的后续研究。