Answers are grounded in this paper and cite their sources.
Demonstrates improved scaling decisions in serverless environments, suggesting enhanced cost efficiency and robustness.
Serverless computing offers automatic resource management and pay-per-use execution, but autoscaling remains difficult due to cold-start latency, inter-function dependencies, and highly dynamic workloads. Many existing approaches scale functions independently or rely on a single predictor, which can reduce robustness and cost efficiency. We present a dependency-aware autoscaling framework that unifies bottleneck identification, short-horizon demand forecasting, and cost-aware control in an end-to-end pipeline. We model applications as directed dependency graphs and prioritize high-impact functions using degree centrality. For these bottlenecks, near-term demand is predicted using lightweight supervised models, whose outputs are fused via a performance-weighted probabilistic ensemble inspired by Bayesian model averaging to improve stability under workload variability. The controller also accounts for cold starts and filters candidate actions through a cost-comparison mechanism to balance latency and operational efficiency. Experiments on real workload traces show improved prediction accuracy and more stable scaling decisions than representative baselines; supervised forecasting also consistently outperforms unsupervised clustering for generating autoscaling actions. The primary contribution is a practical system-level design that integrates dependency analysis, ensemble-based prediction, and cost-aware decision-making for robust serverless autoscaling.
No takes yet. Share an insight, caveat, or question.
Kashaniyan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: