Moirai: Optimizing Quantum Serverless Function Orchestration via Device Allocation and Circuit Deployment

In the rapidly evolving field of quantum computing, the emergence of cloud-based quantum services has opened up access to this cutting-edge technology for a broader user base. As demand for these services grows, the efficient and timely delivery of quantum computing capabilities becomes increasingly critical. However, current queuing methods for executing quantum tasks on specific machines often lead to inefficient resource usage and longer waiting times. In this paper, we present Moirai, a quantum serverless function orchestration framework to reduce the maximum completion time and improve device utilization. Leveraging the unique properties of quantum circuits, Moirai introduces a customized circuit representation scheme, incorporating Directed Acyclic Graph (DAG) conversion and Graph Convolutional Network (GCN) embedding vector feature extraction. Central to Moirai is the use of reinforcement learning, which drives a suite of decision pipelines for device allocation and circuit deployment strategy. We conduct extensive evaluations involving 95 prevalent quantum algorithm circuits across 3∼7 qubits, utilizing 10 IBM quantum physical devices with three types of chip topologies. The results demonstrate that our proposed framework significantly enhances resource allocation efficiency and overall performance, achieving a noteworthy reduction in execution time of over 30% compared to baseline methods. This underscores the effect of Moirai as a practical solution for optimizing quantum cloud services and unlocking the potential of quantum computing technology.