Worker Nodes use

This guide is intended as a comprehensive resource that provides both an overview and a practical introduction to using the SLURM job scheduler on the GRAVITON cluster.

In high-performance computing (HPC), efficient job scheduling is critical to maximizing resource usage and throughput. GRAVITON employs SLURM (Simple Linux Utility for Resource Management), a highly scalable and widely adopted open-source workload manager designed to allocate compute resources, manage job queues, and coordinate the execution of batch and parallel jobs across large clusters.

SLURM is particularly well-suited for scientific workloads requiring fine-grained control over resource allocation, parallel execution, and performance monitoring. It offers users powerful tools to submit, monitor, and manage jobs effectively, while providing system administrators with robust control over scheduling policies and cluster usage.

This documentation covers the key commands, job submission techniques, and best practices for working with SLURM on GRAVITON, from simple batch jobs to multi-node parallel workloads.

Contents:

• Resource Architecture and QoS Policies
  • User Accounts and Groups
  • Partitions and QoS Mapping
  • Partition Architecture
  • Policy Enforcement
• Memory Policy
  • Default Memory Allocation
  • Requesting Double Memory
  • Important Notes
• Submission Evaluation Logic (Lua-based)
  • Key Rules Enforced
  • Job Submission Flow Diagram
• Basic SLURM Commands
  • Job Monitoring
  • Cluster Status
  • Job Submission
  • Queue Information
• Job Submission Policy on GRAVITON
  • Opciones de Slurm prohibidas en GRAVITON
  • Explanation of Allowed Directives
• Non-MPI Jobs
  • Recommended QOS
  • Example 1: Multiple Independent Tasks
  • Example 2: Single Task Using Multiple CPU Cores (Multithreaded or OpenMP)
  • Example 3: Double memory
• MPI Jobs on GRAVITON
  • Partition and QOS Mapping
  • Example 1: MPI Job on a Single Node
  • Example 2: MPI Job on Multiple Nodes
  • Guidelines
  • Using srun vs mpirun