An Auto-Scaling Approach for Serverless Environments Based on a Multi-Expert Consensus Mechanism

This paper contributes to research on serverless computing, auto-scaling, cloud resource management, and workload forecasting by introducing a dependency-aware auto-scaling framework based on a multi-expert consensus mechanism.

Unlike traditional serverless auto-scaling methods that scale functions independently or rely on a single prediction model, this work considers inter-function dependencies in serverless applications. It models applications as directed dependency graphs, identifies high-impact bottleneck functions using degree centrality, and applies short-horizon demand forecasting to support more reliable scaling decisions.

The proposed framework combines multiple lightweight predictive models through a performance-weighted ensemble inspired by Bayesian model averaging. This improves robustness under dynamic workloads while supporting cost-aware resource allocation and reducing the impact of cold-start latency in Function-as-a-Service environments.

Keywords: serverless computing, serverless auto-scaling, Function-as-a-Service, cloud resource management, workload forecasting, dependency-aware scaling, multi-expert consensus, ensemble learning, Bayesian model averaging, cold-start latency, distributed systems, cost-aware optimization.

Citation:

Kashaniyan, M., Ashtiani, M., & Ghassemi, A. (2026). An auto-scaling approach for serverless environments based on a multi-expert consensus mechanism. Journal of Ambient Intelligence and Smart Environments. https://doi.org/10.1177/18761364261459585