ORNL Quantum Computing
The blueprint for a smooth quantum-HPC integration is revealed by Oak Ridge National Laboratory.
Researchers at Oak Ridge National Laboratory (ORNL) of the U.S. Department of Energy have revealed a complete software stack architecture intended to smoothly combine High-Performance Computing (HPC) settings with quantum computing (QC) capabilities. This innovative study, which is described in a publication released by ORNL, offers a basic blueprint for future scientific discovery by proposing a hardware-agnostic framework that addresses major technological obstacles in integrating these two computational paradigms.
In the same way that Graphics Processing Unit (GPU) improve conventional computing through heterogeneous integration, quantum computing is anticipated to act as a specialized accelerator for jobs that are specifically suited to its capabilities rather than to replace it. QPUs are intended to speed up particular quantum-advantaged algorithms inside broader scientific processes, much how GPUs have revolutionized systems like ORNL’s Frontier supercomputer, allowing it to breach the exascale barrier. A strong integration plan with traditional HPC systems is therefore required.
The current study builds on earlier ORNL efforts by emphasising the practical aspects of software architecture and design and offering more specific implementation guidelines. The following are some of the suggested architecture’s key innovations:
- Effective coordination of both quantum and classical resources through a single resource management system.
- A versatile Quantum Programming Interface (QPI) that allows application developers to abstract away hardware-specific specifics.
- An API for the Quantum Platform Manager (QPM) that makes it easier to integrate various quantum hardware systems.
- A full suite of tools for optimizing and implementing quantum circuits.
- A quantum gateway interface for interacting with numerous quantum hardware systems that reduces workload and avoids over-subscription.
In addition to supporting both present-day noisy intermediate-scale quantum (NISQ) devices and upcoming fault-tolerant quantum computers, the framework is made to be hardware-agnostic and compatible with current HPC operations.
The design takes inspiration from the effective incorporation of GPUs into traditional computing. Compilers that pre-process code, divide it into host and quantum segments, and optimize quantum operations into an intermediate representation (IR) are necessary for quantum applications, much like they are for GPU programming. Just-In-Time (JIT) compilation during runtime, similar to GPUs, is one possible hardware-specific optimisation made possible by this quantum-enabled compilation cycle.
The QC/HPC software stack relies heavily on efficient management, which seeks to strike a compromise between increasing application productivity and optimizing resource usage. Based on their quantum and classical resource demands, the study distinguishes three main application patterns: about equal utilization, low quantum/high classical, and high quantum/low classical. Resource allocation must be optimized because quantum gear is scarce, especially when compared to traditional HPC infrastructure.
The framework allows for both interleaved allocations, which allow discrete reservations that may overlap or form a linked series, and simultaneous allocations, which allow quantum and classical computing resources to be reserved simultaneously for the same amount of time. ORNL is investigating how to handle HPC and quantum resources simultaneously by utilizing SLURM’s heterogeneous job (hetjob) feature. For instance, a sbatch script can ask for one QC node for the quantum part and ten nodes for the HPC part. In order to avoid HPC resources from sitting idle while awaiting quantum findings, a credit system is also intended to offer “soft allocations” and quality of service guarantees for quantum workloads.
In order to ensure proper abstraction, the suggested software stack is arranged into discrete levels. The Quantum Programming Interface (QPI) includes APIs for initialization, resource management, execution control, tool configuration, device management, and result processing. The Quantum Platform Manager (QPM), a hardware abstraction layer, simplifies quantum task submission, results retrieval, and device status enquiries. Because of the QPM’s plugin architecture, different suppliers of quantum hardware can create their own plugins and use shared scheduling and communication tools.
The Quantum Toolchain API, which formalises the interface for tools for quantum circuit transformation, is an essential part. Through optimisations like gate reduction and circuit cutting, these tools take quantum programs in formats like QIR or OpenQASM and provide a polished, hardware-compatible representation. These tools can even perform computationally demanding changes by utilising HPC resources.
Using a NWQ-Sim backend, a hybrid quantum-classical application called a variational quantum linear solver (VQLS) was implemented to evaluate the architecture. The framework’s capacity to handle intricate hybrid workflows was demonstrated in this hands-on exercise, which also offered insightful information about application inefficiencies including how optimizer selection affects circuit assessments. Additionally, the framework encourages applications to design for parallel execution by assisting in the identification of any latency problems brought on by sequential circuit construction.
The software stack is made to integrate easily with workflow orchestration frameworks such as Pilot-Quantum in order to guarantee strong integration. Additionally, while upholding operational security and trust boundaries, OLCF’s Secure Scientific Service Mesh (S3M) offers a fundamental infrastructure for the safe access and integration of quantum resources within the larger HPC environment. Comprehensive telemetry capabilities help with real-time monitoring and historical analysis for performance optimisation by gathering operational parameters from both quantum and classical.
Additionally, the framework offers a versatile and scalable modelling environment, acknowledging that quantum computers will continue to be a limited. Through the implementation of a QPM plugin, this environment incorporates different simulator backends, including TNQVM and NWQ-Sim, enabling researchers to test and debug applications on traditional HPC nodes before to deploying them on real quantum hardware.
An excellent foundation for incorporating quantum computing into current HPC infrastructures is offered by this ORNL effort. It tackles important issues with latency, resource management, and workflow optimisation, opening the door for a time when quantum computing would be easily used to speed up computational and scientific processes, eventually opening up new avenues for study and beyond.
With the ultimate goal of greatly accelerating scientific discovery, the long-term goal of combining modelling and simulation, artificial intelligence, and quantum computing into strong, adaptable tools poses a formidable challenge to the computational science community. A crucial step towards the future is provided by this blueprint.
