Title

Kepler Watt Store: Kepler Software Watt Watcher Store

Efficient power utilization has become pivotal in modern software development and deployment strategies. This abstract presents the comprehensive framework of Kepler Watt Store, a set of Cloud Native power-aware tools that incorporate Kepler and Peaks at various stages to monitor and optimize power efficiency throughout the software lifecycle, from development to deployment. Two tools proposed in this abstract, Runtime Monitoring and Power-Efficient Scheduling, are based on existing open-source projects—Kepler and Peaks—and are driven by community collaborations. Continuous Integration and Development Phase are research ideas that we are exploring for their viability in integrating with current software development and CI infrastructure. Some tools in the Kepler Watt Store focus on software development, while others target infrastructure. On the infrastructure side, we propose an operator-based approach to monitor infrastructure power consumption, ensuring backward compatibility for existing infrastructure and future-proofing for new software.

Kepler serves as the cornerstone of this framework, functioning as a sophisticated power observability framework. It is a Kubernetes-based Efficient Power Level Exporter that estimates power consumption at the process, container, and Kubernetes pod levels. The architecture is designed to be extensible, allowing industrial and research projects to contribute novel power models for diverse system architectures. In more detail, Kepler utilizes a BPF program integrated into the kernel’s pathway to extract process-related resource utilization metrics. Kepler also collects real-time power consumption metrics from the node components using various APIs, such as Intel Running Average Power Limit (RAPL) for CPU and DRAM power, NVIDIA Management Library (NVML) for GPU power, Advanced Configuration and Power Interface (ACPI) for platform power, Redfish/Intelligent Power Management Interface (IPMI) also for platform power, or Regression-based Trained Power Models when no real-time power metrics are available in the system.

The abstract emphasizes the prevailing focus of current Kubernetes schedulers on singular objectives, neglecting power efficiency. To fill this void, PEAKS (Power Efficiency Aware Kubernetes Scheduler) emerges, targeting the optimization of aggregate power consumption during scheduling. PEAKS leverages pre-trained Machine Learning models that consider Node Utilization vs Power Consumption to intelligently forecast optimal nodes for scheduling, factoring in workload resource demands. Benchmark assessments reveal heightened power usage in underutilized nodes. PEAKS dynamically situates nodes along the utilization-power curve in real-time, adeptly choosing the best fit according to workload requirements. This proactive approach results in significant power reduction compared to default schedulers, introducing an innovative paradigm in Kubernetes scheduling, emphasizing both multi-objective optimization and power efficiency, thereby contributing to the evolution of cloud-native system management.

The abstract highlights the proactive role of Kepler in CI pipelines for real-time power monitoring during CI runs and its potential integration with Red Hat’s Quarkus framework for Power Observability in the Development Phase. At runtime, Kepler actively monitors power consumption, while Peaks, integrated with Kepler, facilitates power-aware workload allocation, optimizing cluster-wide power consumption during execution.

Continuous Integration (CI): Kepler can be integrated into CI pipelines to monitor power consumption during CI runs. This real-time insight enables early detection of power inefficiencies, fostering proactive adjustments during software integration.

Development Phase: Kepler’s Power Observability framework can measure power readings for software components during the developmental stage. This foresight empowers developers to preemptively optimize software architecture and code to align with power-efficient strategies. We are working on prototyping this at the time of writing this abstract on how such a warning system could integrate with Red Hat’s Quarkus framework.

Runtime Monitoring: Kepler continues its role at runtime, actively monitoring and evaluating real-time power consumption during software execution. This ongoing assessment provides crucial data for immediate optimization and informs future software iterations. We will demonstrate how Kepler provides Power Observability for running clusters.

Power-Efficient Scheduling: Integration of Peaks with Kepler enriches the software lifecycle by enabling power-aware scheduling. Peaks leverages Kepler’s monitoring insights to dynamically allocate workloads, optimizing power consumption across the cluster during execution.

This holistic approach to monitoring power, utilizing Kepler and Peaks, establishes a proactive and sustainable strategy. By addressing power efficiency from inception to execution, this framework drives a paradigm shift towards environmentally conscious and resource-efficient software systems.