Technical Papers Archive

End-to-End Modeling and Trace-Driven Replay for Hybrid Quantum-Classical Supercomputing


Authors: Ziming Zhao and Tingting Li (School of Software Technology, Zhejiang University); Zhaoxuan Li (State Key Laboratory of Cyberspace Security Defense, Institute of Information Engineering, CAS); Xiaofei Yue (Beijing Institute of Technology); Jiongchi Yu (Nanyang Technological University, Singapore); and Jianwei Yin (School of Software Technology, Zhejiang University)

Abstract: Hybrid quantum-classical applications increasingly execute as multi-stage workflows spanning HPC nodes, simulators, workflow runtimes, network services, and remote quantum backends. Evaluating such applications is difficult because their end-to-end critical path crosses heterogeneous components with incompatible observability and control interfaces, making reported results hard to reproduce, compare, and extrapolate. We present QTrace, an end-to-end methodology for performance analysis of hybrid quantum-classical workloads. QTrace combines a unified cross-layer tracing schema with a dependency-aware, trace-driven replay engine that preserves workflow causality while enabling controlled perturbation of backend latency, scheduling policy, and reliability-related effects. This supports both faithful reproduction of observed executions and systematic "what-if" studies for future quantum-accelerated systems. We evaluate QTrace on representative iterative and adaptive workloads across simulator-backed and remote-backend settings. The results show that QTrace reconstructs workflow behavior with high fidelity, reproduces observed executions with low replay error, and exposes bottlenecks that are often hidden by backend-centric metrics alone.



Back to Technical Papers Archive Listing