Joint resource autoscaling and request scheduling for serverless edge computing

Joint resource autoscaling and request scheduling for serverless edge computing

연구 분야: Networking

학회: Cluster Computing

Integrating serverless and edge computing has revolutionized the deployment of software functions on edge devices with limited resources. Nevertheless, the limitations of computing resources, the dynamic nature of users’ requests, and the latency constraints of serverless functions present several challenges for achieving operational serverless edge computing. Motivated by these challenges, this paper proposes a self-adaptive approach through joint resource autoscaling and request scheduling in serverless edge computing environments. Our contributions are two-fold: (1) utilizing a Q-learning algorithm with a customized reward function to dynamically scale the number of function instances handling user requests, and (2) employing a heuristic algorithm to select the warmest instance for scheduling requests among active instances. The proposed Q-learning mechanism considers both workload and resource dynamics to determine the optimal number of active instances. Additionally, the heuristic algorithm improves the performance of Q-learning by choosing the warmest function instance for each incoming request, ensuring minimal instance usage. This research aims to achieve a win–win outcome, benefiting both users and providers. Extensive experiments demonstrate that the proposed approach outperforms baseline methods in terms of request success rate, minimizing utilized instances, and reducing cold start occurrences.