Deadline-Aware Scheduling of Distributed Quantum Circuits in Near-Term Quantum Cloud

TL;DR

Problem: Near-term quantum clouds require LOCC-based distribution and circuit cutting to run large circuits across heterogeneous QPUs while meeting user deadlines. Approach: a deadline-aware scheduling framework that optimizes subcircuit shot distribution and LOCC dependencies, solved via simulated annealing. Contributions: first unified model combining deadlines, LOCC wire-cut dependencies, and shot distribution; O1 optimization; extensive simulations showing improved on-time service and shorter makespans versus baselines. Significance: enables scalable, deadline-driven DQC for multi-user LOCC-only clouds on heterogeneous QPUs.

Abstract

Distributed quantum computing (DQC) enables scalable quantum computations by distributing large quantum circuits on multiple quantum processing units (QPUs) in the quantum cloud. In DQC, after partitioning quantum circuits, they must be scheduled and executed on heterogenous QPUs while balancing latency, overhead, QPU communication resource limits. However, since fully functioning quantum communication networks have not been realized yet, near-term quantum clouds will only rely on local operations and classical communication settings between QPUs, without entangled quantum links. Additionally, existing DQC scheduling frameworks do not account for user-defined execution deadlines and adopt inefficient wire cutting techniques. Accordingly, in this work, a deadline aware DQC scheduling framework with efficient wire cutting for near-term quantum cloud is proposed. The proposed framework schedules partitioned quantum subcircuits while accounting for circuit deadlines and QPU capacity limits. It also captures dependencies between partitioned subcircuits and distributes the execution of the sampling shots on different QPUs to have efficient wire cutting and faster execution. In this regard, a deadline-aware circuit scheduling optimization problem is formulated, and solved using simulated annealing. Simulation results show a marked improvement over existing shot-agnostic frameworks under urgent deadlines, reaching a 12.8% increase in requests served before their deadlines. Additionally, the proposed framework serves 8.16% more requests, on average, compared to state-of-the-art dependency-agnostic baseline frameworks, and by 9.60% versus the dependency-and-shot-agnostic baseline, all while achieving a smaller makespan of the DQC execution. Moreover, the proposed framework serves 23.7%, 24.5%, and 25.38% more requests compared to greedy, list scheduling, and random schedulers, respectively.

Figures (5)

• Figure 1: End-to-end near-term quantum cloud DQC workflow. User circuits with deadlines enter a request buffer, are cut into subcircuits by the compiler, scheduled on eligible QPUs under qubit count, depth and precedence constraints. Then, final results are reconstructed with classical post-processing.
• Figure 2: Average served requests vs. Number of requests.
• Figure 3: Average makespan vs number of requests.
• Figure 4: Served requests rate vs deadline factor and number of requests.
• Figure 5: Scheduling outputs of different benchmarks.