Kubernetes Cloud-Edge CPU and Memory Utilization Time Series

280.23 hours of time-series data from controlled Kubernetes experiments, comprising 250,432 timestamped samples collected between May and September 2025. The dataset captures worker-level CPU and memory utilization under dynamic workloads across multiple Kubernetes distributions and network interfaces.

Use Cases

• Train time-series models on 30-second sampled CPU and memory utilization data to forecast resource demand under pod-scaling scenarios.
• Analyze the impact of Kubernetes distributions (Vanilla Kubernetes, K3s) and CNI plugins (Flannel, Calico) on resource usage patterns during mixed workload experiments.
• Detect workload transitions using data from two-stage mixed workload experiments that feature mid-run shifts in pod count and stress parameters.
• Benchmark resource consumption for stable-low, brief-high, and long-medium workload profiles to model realistic cloud-edge service loads.

Strengths

• 250,432 timestamped samples provide a substantial volume of time-series observations.
• Data spans 280.23 hours (~12 days) across three distinct experimental scenarios.
• Incorporates heterogeneity from multiple Kubernetes distributions and network interfaces.
• Metrics are sampled at a consistent 30-second interval for continuous evaluation.

Limitations

• Limited to three specific experimental scenarios, which may not cover all real-world deployment conditions.
• Data is generated from controlled benchmarking experiments, not from a live production environment.
• The 0.0 B size indicates a tiny dataset, suggesting limited scope for large-scale model training.

Provenance

Source

George Koukis via figshare.

Collection Method

Generated using the CODEF experimentation framework with stress-ng and kube-burner tools on Kubernetes-based infrastructures.

Time Range

May to September 2025.

Freshness

Data collected between May and September 2025; last updated March 2026.

Geography

null