EUVIMEDEUVIMED entdecken
Crossref

Adaptive Function Service Auto-Scaling for Serverless Computing via Deep Recurrent Reinforcement Learning

Yueshen Xu, Guoliang Mi, Qingshan Li, Jianwei Yin, Tom H. Luan, Wei Shao, Rui Li

IEEE Transactions on Services Computing · 2026-07 · 19 (4) · 2674-2687

DOI: 10.1109/tsc.2026.3710828 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Services Computing, 2026-07. Verfasst von Yueshen Xu und 6 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: National Key Research and Development Program of China, National Natural Science Foundation of China, Open Foundation of Yunnan Key Laboratory of Software Engineering.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As an emerging service provision paradigm, Function-as-a-Service (FaaS) can automatically manage function deployment and resource scaling, thereby reducing operational overhead, and has experienced rapid development. A core goal of FaaS is providing on-demand elastic scaling, but existing FaaS platforms usually rely on fixed-threshold-based scaling approaches, which lead to several serious issues. For example, the setting of the threshold depends heavily on engineers’ expertise, and it is difficult to handle those unexpected situations such as dynamic workloads and unpredictable demands. Some other approaches suffer from the ignorance of several performance metrics. To address these issues, in this article, we develop a novel adaptive auto-scaling approach for function services. Our approach comprehensively studies a set of performance metrics, including CPU, memory, throughput, latency, and function instance replicas. So our approach is empowered with the ability of fully exploring the characteristics of cloud workloads, especially under dynamic conditions. We provide a series of computation methods for all metrics and develop a multi-objective optimization model. We further propose a tailored optimization algorithm that aims to address the partial observability issue commonly existing in function services. We enhance Proximal Policy Optimization (PPO) by introducing RNNs and develop a Deep Recurrent Reinforcement Learning (DRRL) algorithm, in which we also design a novel multi-metric reward function. We conducted sufficient experiments on the open-source FaaS platform OpenFaaS and employed a real-world dataset released by Microsoft Azure. The results show that our approach consistently achieves superior performances across all metrics compared to a group of strong baselines, including the method employed in OpenFaaS.

Förderung

  • National Key Research and Development Program of China
  • National Natural Science Foundation of China
  • Open Foundation of Yunnan Key Laboratory of Software Engineering

Zitieren

APA

Xu, Y., Mi, G., Li, Q., Yin, J., Luan, T. H., Shao, W., & Li, R. (2026). Adaptive Function Service Auto-Scaling for Serverless Computing via Deep Recurrent Reinforcement Learning. IEEE Transactions on Services Computing, 19(4), 2674-2687. https://doi.org/10.1109/tsc.2026.3710828

Vancouver

Xu Y, Mi G, Li Q, Yin J, Luan TH, Shao W, et al. Adaptive Function Service Auto-Scaling for Serverless Computing via Deep Recurrent Reinforcement Learning. IEEE Transactions on Services Computing. 2026;19(4):2674-2687. doi:10.1109/tsc.2026.3710828

BibTeX
@article{Xu2026,
  author = {Xu, Yueshen and Mi, Guoliang and Li, Qingshan and Yin, Jianwei and Luan, Tom H. and Shao, Wei and Li, Rui},
  title = {Adaptive Function Service Auto-Scaling for Serverless Computing via Deep Recurrent Reinforcement Learning},
  journal = {IEEE Transactions on Services Computing},
  year = {2026},
  volume = {19},
  number = {4},
  pages = {2674--2687},
  doi = {10.1109/tsc.2026.3710828}
}
Kostenlos starten →