Overview
Efficient use of compute nodes ensures fair access for all users and maximizes throughput — proper resource management prevents waste and benefits the whole community.
Recommendations
- Use shared queues whenever feasible so multiple jobs can run on the same nodes
- Request only what you need — cores, nodes, memory, and GPUs your job will actually use
- Allocate full nodes for interactive sessions only when all cores will be used; otherwise pick a shared queue to avoid idle cores
- Monitor consumption — track memory, CPU, and GPU usage (see managing jobs)
- Optimize code — faster completion frees resources for others
- Release nodes promptly after job completion
Summary
Prioritize shared queues, monitor jobs closely, and review the queue tables before submission. These practices work alongside fairshare and job priority to keep workflows predictable and efficient.
Applies to