Background and Motivation
Edge computing has emerged as a critical paradigm for reducing latency and bandwidth usage by processing data closer to users and IoT devices. However, the proliferation of edge deployments introduces significant energy consumption challenges, particularly in resource-constrained environments where power efficiency directly impacts operational costs and environmental sustainability.
Modern container orchestration frameworks like Kubernetes have revolutionized application deployment and management, but they lack comprehensive energy awareness in their scheduling and resource allocation decisions. Current orchestration systems primarily optimize for performance metrics such as CPU utilization and memory consumption while remaining largely oblivious to energy consumption patterns. This gap becomes critical in edge environments where devices operate on limited power budgets and energy efficiency directly impacts battery life and cooling costs.
Kepler (Kubernetes-based Efficient Power Level Exporter) represents a breakthrough in container-level energy monitoring, using eBPF to probe CPU performance counters and system statistics to estimate workload energy consumption. However, existing orchestration frameworks do not effectively leverage such energy metrics for intelligent scheduling and resource management decisions. The challenge lies in integrating fine-grained energy accounting into orchestration logic to enable energy-aware scheduling, resource allocation, and workload placement across heterogeneous edge hardware.
Expected Outcomes
This thesis aims to develop an energy-aware orchestration framework that automatically optimizes edge deployments for minimal energy consumption while maintaining performance requirements. The research will begin by conducting comprehensive energy profiling of modern orchestration operations, quantifying the energy costs of container lifecycle management, service discovery, load balancing, and inter-service communication across diverse edge hardware platforms.
The core contribution will be the design and implementation of energy-aware scheduling algorithms that leverage real-time energy metrics from Kepler to make intelligent placement decisions. The system will incorporate energy consumption models for different workload types and hardware configurations, enabling predictive energy optimization during deployment planning. The framework will support dynamic workload migration based on energy efficiency metrics and hardware utilization patterns.
Integration with existing orchestration platforms will be achieved through custom schedulers and resource managers that embed energy accounting into decision-making processes. The system will provide energy budgeting capabilities, allowing operators to set energy consumption limits and automatically optimize deployments to stay within those constraints. Real-time energy monitoring dashboards will provide visibility into energy consumption patterns at container, pod, and cluster levels.
Experimental evaluation will compare energy consumption across different scheduling strategies using both synthetic workloads and real-world edge applications. The research will quantify energy savings achieved through intelligent scheduling while measuring the impact on application performance metrics such as latency and throughput. The study will establish best practices for energy-efficient edge orchestration and provide guidelines for operators seeking to minimize their environmental footprint.
Requirements
• Strong understanding of container orchestration systems and Kubernetes architecture • Experience with eBPF programming • Proficiency in Go programming for Kubernetes controller and scheduler development • Familiarity with energy measurement tools and power profiling techniques
Related Work
[1] Jiang, C., Fan, T., Gao, H., Shi, W., Liu, L., Cérin, C., & Wan, J. (2020). Energy aware edge computing: A survey. Computer Communications, 151, 556-580.
[2] Bartolomeo, Giovanni, et al. "Oakestra: A lightweight hierarchical orchestration framework for edge computing." 2023 USENIX Annual Technical Conference (USENIX ATC 23). 2023.
