Why Dependency-Aware Auto-Scaling Matters in Serverless Computing

Serverless computing provides automatic resource management and pay-per-use execution, but efficient auto-scaling remains challenging. Dynamic workloads, cold-start latency, and inter-function dependencies can make scaling decisions unreliable when each function is treated independently.

In my paper, “An Auto-Scaling Approach for Serverless Environments Based on a Multi-Expert Consensus Mechanism,” we propose a dependency-aware auto-scaling framework for Function-as-a-Service environments.

The framework models serverless applications as directed dependency graphs, identifies high-impact bottleneck functions using degree centrality, and applies short-horizon workload forecasting. Instead of relying on one predictor, the system combines multiple lightweight predictive models using a multi-expert consensus mechanism inspired by Bayesian model averaging.

This research contributes to serverless computing, cloud resource management, workload forecasting, dependency-aware scaling, ensemble learning, and cost-aware optimization.

Paper

An Auto-Scaling Approach for Serverless Environments Based on a Multi-Expert Consensus Mechanism
Mobina Kashaniyan, Mehrdad Ashtiani, Amirhossein Ghassemi
Published in Journal of Ambient Intelligence and Smart Environments, SAGE Publications.

DOI: https://doi.org/10.1177/18761364261459585

Paper page: https://mobinakashaniyan.github.io/publications/consensus-autoscaling-serverless/

PDF: https://mobinakashaniyan.github.io/papers/an-auto-scaling-approach-serverless-environments-multi-expert-consensus.pdf