Power-Aware CPU Cap Mechanism in Serverless Computing Environments

Designing elastic resource allocation algorithms for serverless environments is challenging but promising, as slight performance improvements yield significant monetary savings for service providers. Serverless computing supports execution of pipeline operations without explicit resource provisioning. However, effective consolidation strategies are needed to minimize contention among concurrently running applications and to handle workload surges. This article presents CPU cap strategies for optimizing power consumption by adjusting CPU voltage and frequency, while adhering to user-defined latency event-driven applications in serverless platforms. Our strategy reduces power consumption by 28% for high-priority applications in a serverless environment managed by OpenWhisk, while maintaining the 99th percentile latency within the requested range.