Optimize power consumption for serverless computing

Abstract

This project investigates the optimization of power consumption in serverless computing environments through the integration of the Kubernetes (K8s) Power Manager into vHive, a full-stack open-source ecosystem for serverless cloud benchmarking, experimentation, and innovation. The primary goal is to verify the efficiency of the K8s Power Manager in reducing power consumption without impeding performance. A series of experiments were conducted to evaluate the impact of various power management strategies on workload latency and power consumption. These experiments include workload sensitivity analysis, internode scaling, and controlled versus uncontrolled assignment of workloads to nodes based on their frequency sensitivity. The results demonstrate that adjusting CPU frequency according to workload sensitivity can lead to significant energy savings while maintaining or improving latency. Specifically, a balanced configuration of high and low-frequency nodes and assigning workloads to nodes based on their frequency sensitivity were found to optimize power consumption and performance. The study concludes that the K8s Power Manager, when implemented within a well-defined policy framework, can significantly enhance power optimization in serverless computing environments, leading to a more sustainable and cost-effective cloud infrastructure.