ML-Based Performance Modeling in Edge FaaS Systems

The Function as a Service (FaaS) model is becoming increasingly attractive for both cloud and edge computing scenarios, offering a solution where self-contained functions are executed in response to specific events. In this model, the complexities of load balancing and scaling are handled by the service providers. However, accurate resource consumption estimates are crucial in FaaS-enabled clusters, especially at the edge, to optimize resource efficiency, minimize latency, prevent overloads, and ensure scalability. This work focuses on performance modeling within FaaS-enabled distributed and decentralized edge environments, at both the node and function levels. Using a Machine Learning (ML)-based approach, we propose a framework to predict key performance metrics, such as CPU usage, memory, and energy consumption. Additionally, we forecast potential system overloads based on the incoming load. By introducing a profiling tool that characterizes functions by their resource usage patterns, prediction of node-level resource consumption without needing detailed function-level knowledge is made possible. Experimental results show that our models achieve 97% accuracy in predicting node overloads.