Slurm 26.11 introduces HRES plugins, allowing quantum processors to be scheduled as native resources in supercomputing centers.
Quantum processing units (QPUs) are being integrated into supercomputing centers globally, driven by national programs in North America, Europe, and Asia. This marks a significant step toward hybrid quantum-classical supercomputers, which are expected to redefine scientific research capabilities in fields such as chemistry, weather modeling, and engineering simulations. To facilitate this transition, Slurm 26.11 introduces Hierarchical Resources (HRES) plugins, enabling QPUs to be scheduled as native, cluster-level resources.
Until now, integrating QPUs into high-performance computing (HPC) systems has been hindered by the lack of a standardized workload management solution. Existing methods, such as using generic resources (GRES) or license-based scheduling, have proven inadequate. GRES ties QPUs to individual compute nodes, limiting flexibility and risking downtime if a node fails. License-based approaches, while cluster-wide, can leave compute nodes idle while waiting for QPU availability. Additionally, these methods often require custom integrations for each supercomputing site, increasing complexity and maintenance costs for administrators and QPU vendors alike.
11 addresses these challenges with its HRES plugin interface. The first implementation, hres/lua , allows QPUs to be modeled as hierarchical resources. This approach decouples site-specific policies from device logic, enabling a single Lua script to handle device health checks, job admission, and resource release. yaml file, specifying details such as endpoints, node access, and maintenance cycles. This modular design ensures portability across Slurm deployments, reducing integration overhead for QPU vendors and global systems integrators.
Source link







