Qoala: an Application Execution Environment for Quantum Internet Nodes

TL;DR

Qoala advances quantum internet programming by delivering a unified hybrid classical-quantum program format and a runtime task-based representation that enables joint scheduling across classical and quantum resources. Implemented as a modular, hardware-validated simulator on NetSquid, Qoala demonstrates a two-level hierarchical scheduler (node, CPS, QPS) and a flexible program structure that supports asynchronous and interactive execution, entanglement requests, and multitasking. Compared with NetQASM, Qoala enables cross-domain compilation and scheduling improvements, achieving higher application success under realistic hardware constraints and network schedules. The work highlights the critical tradeoffs between classical latency, quantum decoherence, and scheduling policies, and provides an open-source platform to explore advanced compilers and scheduling strategies for quantum internet applications. Overall, Qoala paves the way for practical, platform-agnostic execution of quantum internet workloads and motivates future hardware integration and compiler research.

Abstract

Recently, a first-of-its-kind operating system for programmable quantum network nodes was developed, called QNodeOS. Here, we present an extension of QNodeOS called Qoala, which introduces (1) a unified program format for hybrid interactive classical-quantum programs, providing a well-defined target for compilers, and (2) a runtime representation of a program that allows joint scheduling of the hybrid classical-quantum program, multitasking, and asynchronous program execution. Based on concrete design considerations, we put forward the architecture of Qoala, including the program structure and execution mechanism. We implement Qoala in the form of a modular and extendible simulator that is validated against real-world quantum network hardware (available online). However, Qoala is not meant to be purely a simulator, and implementation is planned on real hardware. We evaluate Qoala's effectiveness and performance sensitivity to latencies and network schedules using an extensive simulation study. Qoala provides a framework that opens the door for future computer science research into quantum network applications, including scheduling algorithms and compilation strategies that can now readily be explored using the framework and tools provided.

Figures (27)

• Figure 1: Example application consisting of two hybrid classical-quantum programs (on Nodes 1 and 2) including (1) Entanglement generation between two qubits (circles) in a synchronized time slot (defined by network controller). (2) A local measurement of qubit A at Node 1 resulting in a classical outcome bit (destroying the qubit) (4) Communication of the classical bit from Node 1 to Node 2 (taking non-deterministic time) (5) Execution of a quantum circuit on qubit B at Node 2 depending on the classical bit. The quality of qubit B has degraded during the time elapsed since (1). (6) Node 2 measures qubit B and outputs the classical result.
• Figure 2: QNodeOS donne2024design vs. Qoala capabilities.
• Figure 3: Minimal hardware assumptions for a single node. A Classical Processing System (CPS) can execute classical code and can communicate classical messages with other nodes in the network. A Quantum Processing System (QPS) can execute quantum code and can realize entanglement (quantum connections) with other nodes in the network. The CPS and QPS are controlled by a scheduler, and have access to shared memory. In the QNodeOS architecture donne2024design, the CPS is realized as the CNPU, and the QPS as the combined QNPU-QDevice system. For Qoala, we only focus on the classical-quantum distinction, and not the internal implementation (such as a QNPU-QDevice separation), hence the different terminology.
• Figure 4: High-level overview of Qoala: An SDK allows program code in a high-level language (e.g. Python). A compiler translates this code into a Qoala program (specific compiler not in scope of this work). To run, a program is instantiated with concrete values for program arguments. Tasks are created for the program instance, which are scheduled and executed by the scheduler. Multiple program instances may exist at the same time (both multiple instances of the same or different programs). All tasks from all instances are added to a single task graph (\ref{['sec:tasks']}) used by the scheduler.
• Figure 5: Example Qoala program containing a host section with 4 blocks, a local routine (subrt1), and a request routine (req1). Block b2 has a relative deadline to b0 of $0.1$ times qubit noise parameter $T_2$.