Performance Analysis of QAOA Across Distributed Quantum Network Topologies Using SwitchQNet

As distributed quantum computing (DQC) architectures move toward modular quantum data centers (QDCs), there is a critical need to understand how specific algorithmic workloads interact with network-level constraints. This paper investigates the performance of the Quantum Approximate Optimization Algorithm (QAOA) when executed across distributed quantum networks, specifically focusing on how its graph-based communication requirements impact latency and resource overhead in switch-based architectures.

The researchers extended the SwitchQNet compiler framework—a tool designed to manage entanglement generation and routing in QDCs—to support QAOA. They developed a routing generator that translates the two-qubit cost interactions of a QAOA problem graph into remote-CX communication requests. The team then evaluated this workload across three common network topologies: Clos, fat-tree, and spine-leaf. The evaluation measured performance metrics including communication latency, EPR-pair overhead, and retry rates, while testing sensitivity to parameters like buffer size, look-ahead depth, and EPR fidelity.

The study found that the SwitchQNet framework consistently provides latency reductions for QAOA, typically improving performance by a factor of 1.9x to 2.2x compared to baseline compilers. These improvements are driven by the framework's ability to hide switch reconfiguration latency through look-ahead scheduling and collective in-rack entanglement generation. The results indicate that while QAOA’s communication demand is dictated by the problem graph structure, the network architecture remains a significant bottleneck, particularly regarding cross-rack communication, which dominates total latency despite representing a smaller fraction of overall entanglement operations.