Basic Usage

• Overview
  • Login Node
  • Compute Nodes
  • Job Queues (Partitions)
• Preparing Job Scripts (sbatch)
  • #SBATCH Directives
  • Basic Options
  • Time Limit
  • CPU Allocation
  • Memory Allocation
  • GPU Allocation
  • Node Allocation
  • Non-restartable Jobs
  • Quality of Service
  • Email Notification
  • Slurm Environment Variables
• Submitting and Managing Jobs
  • Resource Usage Estimation
  • Checking Resource Availability
  • Submitting Batch Jobs
  • Status Information
  • Cancelling Jobs
  • Pending Jobs
• Checking Job Efficiency
  • CPU and memory efficiency
  • Resource usage of running jobs
  • Requested vs allocated resources
  • GPU usage statistics
  • Job monitoring with jobstats
  • Aggregated job efficiency with reportseff
• Troubleshooting Resource Efficiency
  • Start with measurements
  • CPU parallelism and scaling
  • GPU activity and data supply
  • Host memory and GPU device memory
  • Lustre I/O and metadata activity
  • NVLink and communication
  • Runtime requests and interactive allocations
  • Missing measurements and short jobs
• Interactive Use
  • srun
• Exposing Web Services with frp
  • Download frpc (client)
  • Example frpc.toml configuration (HTTP)
  • Slurm example: publish a Jupyter notebook from a job
  • Troubleshooting
  • Security notes
• Resource Usage
  • Balance and Billing
  • Total Consumption
  • Storage Quotas
• Quality of Service (QOS)
  • Checking Available QOSs
  • Default QOS
  • Manual QOS
• Advanced Job Definitions
  • Job Array
  • Packed Jobs
  • Multiple Tasks
• Data Sharing for Project Members
  • ACL
  • Examples
• File Management Considerations
  • Inode Limit
  • FUSE