MC-Kube: Mixed-Criticality Workload Management for Kubernetes

MC-Kube is a Kubernetes operator designed for mixed-criticality real-time workloads (e.g., autonomous driving, industrial IoT). It overcomes the limitations of the default CFS scheduler by integrating a hierarchical-CBS kernel patch and eBPF-based overrun detection, ensuring strong temporal guarantees with minimal overhead.

📄 Paper: SDV에서의 Mixed-Criticality Task Orchestrator (Submitted to HUFS, 2025)

Key Features

• ⚡ Zero-Overhead Monitoring: eBPF-based detection with 246× lower latency (299µs) compared to API polling.
• 🛡️ Strong Isolation: Container-level SCHED_DEADLINE support via custom Kernel patches.
• 🧠 Adaptive Orchestration:
  • Runtime Escalation: Automatically boosts runtime (runtime_low → runtime_hi) upon deadline miss.
  • Smart Preemption: Migrates high-criticality tasks to secure cores by preempting lower-priority pods.
• 📈 Proven Performance: Achieved 86% reduction in deadline misses in high-load scenarios.

Architecture

Quick Start

Prerequisites

• Kubernetes v1.34+ Cluster
• Linux Kernel v6.13+ with hierarchical-CBS patch (REQUIRED)
• kubectl, docker

Installation

Deploy the MC-Kube operator and CRDs in one go:

# 1. Install CRDs & Operator
make install
make deploy IMG=ghcr.io/hufs-ese-lab/mc-kube:v1

# 2. Check Deployment
kubectl get pods -n mc-kube-system

Usage Example

Apply a sample mixed-criticality workload:

kubectl apply -k config/samples/

To verify the real-time configuration applied to the container:

kubectl exec <pod-name> -- cat /sys/fs/cgroup/cpu.rt_multi_runtime_us

Performance & Reproducibility

Our experiments show that MC-Kube significantly outperforms standard kubernetes under heavy loads.

Reproduce the experiments:

# Run scenarios
kubectl apply -f config/samples/test-preemption-scenario.yaml

# Monitor status
kubectl get mckubes -o wide

License

This project is licensed under the Apache 2.0 License.

Research Group: ESE Lab, Hankuk University of Foreign Studies