Orchestrating Function Placement for Serverless Sky Computing

In multi-cloud sky computing environments, optimizing the placement of functions is critical for enhancing the performance of Function-as-a-Service (FaaS) applications. However, existing approaches often struggle due to the large decision space brought by the diversity of computation nodes and varying pricing strategies. In this paper, we propose a novel deep reinforcement learning (DRL)-based online function placement (DOFP) algorithm, which formulates the problem as a Markov decision process (MDP). Specifically, the DOFP algorithm utilizes a D3QN agent to learn from the current state and environmental context. A dynamic ϵ-greedy strategy is adopted to balance exploration and exploitation effectively. Through extensive simulation experiments, we show that DOFP outperforms existing baseline algorithms, achieving faster convergence and greater stability in training, while significantly reducing the cost under latency and resource constraints.