Qonductor: A Cloud Orchestrator for Quantum Computing

TL;DR

This work addresses the challenge of orchestrating hybrid quantum-classical workloads on heterogeneous, resource-scarce quantum clouds. It introduces Qonductor, a hardware-agnostic cloud orchestrator with a hybrid resource estimator and a Pareto-optimal scheduler that balances fidelity and job completion time while improving QPU utilization. The approach is validated with real IBM Quantum data and large-scale circuit benchmarks, achieving up to 54% faster completion times with minimal fidelity loss and substantial resource utilization gains, while providing accurate fidelity and runtime estimates and scalable scheduling. The results demonstrate that combining classical resources for error mitigation with intelligent, multi-objective scheduling can significantly enhance the practicality and efficiency of cloud-based quantum computing.”

Abstract

We describe Qonductor, a cloud orchestrator for hybrid quantum-classical applications that run on heterogeneous hybrid resources. Qonductor abstracts away the complexity of hybrid programming and resource management by exposing the Qonductor API, a high-level and hardware-agnostic API. The resource estimator strategically balances quantum and classical resources to mitigate resource contention and the effects of hardware noise. The hybrid scheduler automates job scheduling on hybrid resources and balances the tradeoff between users' objectives of QoS and the cloud operator's objective of resource efficiency. We implement an open-source prototype and evaluate Qonductor using more than 7000 real quantum runs on the IBM quantum cloud to simulate real cloud workloads. Qonductor achieves up to 54% lower job completion times (JCTs) while sacrificing 3% execution quality, balances the load across QPU, which increases quantum resource utilization by up to 66%, and scales with growing system sizes and loads.

Figures (10)

• Figure 1: Quantum cloud and hybrid computational model (§ \ref{['section:background:quantum-cloud']}). Quantum applications are hybrid, i.e., require quantum and classical resources. The pre-processing, compilation, and post-processing steps run on classical heterogeneous accelerators. QPUs are vastly heterogeneous across space and time, i.e., QPU technologies, architectures, and calibration data.
• Figure 2: Quantum orchestration challenges (§ \ref{['section:motivation:characteristics-nisq']}). (a)Impact of circuit cutting as a relative increase in execution fidelity, quantum, and classical runtime for 12-qubit and 24-qubit circuits. (b)Spatial performance variance: fidelity of a 12-qubit GHZ circuit on different IBM QPUs. There is a 38% fidelity difference from best to worst QPU.(c) QPU load imbalance: number of pending jobs on different IBM QPUs. There is up to $\sim$100$\times$ load difference across QPUs.
• Figure 3: Qonductor overview (§ \ref{['section:overview']}). Qonductor comprises the control plane, data plane, worker node(s), and the system monitor. Core components are highlighted as light green boxes. The control plane performs resource estimation, job management, and hybrid scheduling. The data plane is used to deploy and invoke hybrid images. Workers manage hybrid resources.
• Figure 4: Resource estimator workflow (§ \ref{['section:resource_estimator']}). (a) Error mitigation is applied to the original circuit, which generates one or more circuits. (b) The generated circuits are compiled for template QPUs, generating executable circuits. (c) Fidelity is estimated for all generated circuits where the minimum fidelity binds the solution fidelity. (d) The classical and quantum task execution times are estimated. (e) Resource plans are generated based on the estimated fidelities and total execution times.
• Figure 5: Quantum scheduler workflow (§ \ref{['section:scheduler']}). (a) Job pre-processing: the jobs and QPUs are filtered based on the configuration options. Then, the fidelity and execution time estimations are fetched from the system monitor datastore. (b) Multi-objective optimization: we use the NSGA-II genetic algorithm to create a Pareto front of solutions. (c) Selection: we select one of the solutions based on multiple-criteria decision-making (MCDM) that uses pseudo-weights for fidelity and JCT.