SchedMD/slurm
Slurm
Slurm ist ein Cluster-Workload-Manager und Job-Scheduler für High-Performance-Computing-Umgebungen. Es fungiert als verteilter Compute-Orchestrator, der groß angelegte Rechenaufgaben in einer Warteschlange verwaltet und über mehrere Compute-Nodes in einem Cluster ausführt.
Das System agiert als Ressourcen-Arbitrator, der Hardware-Nodes und Prozessoren unter konkurrierenden Benutzern verteilt, um Ressourcenkonflikte zu vermeiden und die Effizienz zu maximieren. Es koordiniert den gleichzeitigen Start mehrerer Prozesse über verschiedene physische Server hinweg, um parallele Jobs und wissenschaftliche Workloads auszuführen.
Die Plattform deckt breite Aufgabenbereiche ab, einschließlich Batch-Job-Scheduling, Compute-Ressourcen-Allokation und paralleler Workload-Ausführung. Es steuert das Timing und die Ausführung von Jobs basierend auf Ressourcenverfügbarkeit und Priorität.
Features
- HPC Resource Allocation - Manages hardware resource allocation and job priority specifically for high-performance computing environments.
- High-Performance Computing - Executes computationally intensive tasks across distributed hardware architectures using a cluster manager.
- Managed Cluster Batch Execution - Queues and executes large-scale computational batch workloads across multiple compute nodes in a cluster.
- Process Coordination - Synchronizes the simultaneous launch of multiple processes across different physical servers for parallel jobs.
- HPC Cluster Orchestrations - Provides orchestration and resource management of large-scale computational workloads across high-performance computing clusters.
- Cluster Workload Managers - Controls the execution and timing of high performance computing jobs to ensure optimal hardware utilization.
- Priority-Based Job Schedulers - Queues and executes batch computational workloads based on priority and resource availability.
- Batch Schedulers - Queues and executes computational batch tasks based on priority and available hardware.
- Job Execution Controllers - Provides runtime control over the start, stop, and monitoring of work across allocated compute nodes.
- Remote Job Dispatchers - Dispatches control signals from a central manager to specific nodes to trigger task execution.
- Resource Allocation - Distributes hardware nodes and processors across a cluster to ensure efficient utilization.
- Queue Priority Scheduling - Orders pending workloads in a logical queue and dispatches them according to user priorities.
- Parallel Job Orchestrators - Coordinates the simultaneous launch and execution of parallel workloads across multiple processes and distributed compute nodes.
- Distributed Workload Execution - Runs single computational tasks across multiple compute nodes simultaneously to reduce processing time.
- Resource Access Arbitrators - Resolves conflicting requests for available system resources through a centralized coordinator.
- Distributed Node Agents - Uses lightweight agents on every compute node to manage local job execution and report health status.
- Distributed Job Executors - Executes and monitors computational workloads by integrating with distributed compute nodes.
- Scientific Workload Schedulers - Schedules the timing and execution of complex scientific workloads across a network of distributed servers.
- State-Based Resource Tracking - Maintains a real-time database of node availability and allocations to prevent hardware oversubscription.
- Infrastructure and Deployment - Scalable workload manager for high-performance computing.
- Infrastrukturmanagement - Skalierbarer Workload-Manager für High-Performance Computing.
- Resource Scheduling - Highly scalable workload manager for high-performance computing.
