Quantum-inspired bi-level scheduling for resource-aware serverless function deployment in edge computing

Abstract The implementation of software functions on resource-constrained edge devices has been revolutionized by the integration of serverless and edge computing. However, attaining efficient serverless edge computing is complicated by a number of factors, including the latency limits of serverless operations, the dynamic nature of user requests, and the limitations of computing resources. The majority of current strategies rely on heuristic or static scheduling techniques, which are unable to adapt to dynamically changing edge environments and various function execution patterns. Therefore, in this work, we introduced a quantum-inspired bi-level scheduling framework to place the functions in a serverless edge computing environment. Prior to scheduling, we employed the Improved Quantile Regression Neural Network (IQRNN) to predict the execution time, CPU, memory, bandwidth, and storage usage for each function. Afterward, a quantum-inspired Northern Goshawk Optimization algorithm (QNGO) based global scheduler is utilized in the first level of scheduling to select a suitable environment for the function. Then, the Proximal Policy Optimization-based deep reinforcement learning algorithm (PPO-RL) is used in the second level of scheduling to select the suitable container to place the function. Extensive experiments on two real-world datasets, namely the Azure and Globus Compute datasets, demonstrate the effectiveness of the proposed method. Compared with existing techniques, the proposed framework reduces 20.17–29.16% cold start latency, 19.45–20.12% energy consumption, 13.68–16.39% function operating cost and 24.31–27.07% container fault rate.