3. Performance and Cost Optimization

Timing information and scaling plots to assist users in optimizing the performance of their parallel cluster.

• 3.1. ParallelCluster Configuration
• 3.2. CMAQv5.4 Benchmarks
  • 3.2.1. An explanation of why a scaling analysis is required for Multinode or Parallel MPI Codes
• 3.3. Slurm Compute Node Provisioning
• 3.4. Benchmark Timings for CMAQv5.4 12US1 Benchmark
  • 3.4.1. Benchmark Timing for c6a.48xlarge
  • 3.4.2. Benchmark Timing for hpc6a.48xlarge
  • 3.4.3. Benchmark Timing for hpc7g.8xlarge with 32 processors per node
  • 3.4.4. Benchmark Timing for hpc7g.16xlarge with 64 processors per node
• 3.5. Benchmark Scaling Plots for CMAQv5.4 12US1 Benchmark
  • 3.5.1. Benchmark Scaling Plot for hpc6a.48xlarge
  • 3.5.2. Benchmark Scaling Plot for hpc7g.8xlarge
  • 3.5.3. Benchmark Scaling Plot for hpc7g.16xlarge
  • 3.5.4. Total Time and Cost versus CPU Plot for hpc7g.8xlarge
• 3.6. Cost Information
  • 3.6.1. Cost Explorer
  • 3.6.2. Compute Node Cost Estimate
  • 3.6.3. Storage Cost Estimate
  • 3.6.4. Annual simulation local storage cost estimate
  • 3.6.5. Archive Storage cost estimate for annual simulation - assuming you want to save it for 1 year
  • 3.6.6. CycleCloud and ParallelCluster Price Comparison of Cost Estimate for Annual Simulation (Filesystem + Compute)
  • 3.6.7. Assumptions for Price Estimate for Annual Simulation (Filesystem + Compute)
• 3.7. Recommended Workflow for extending to annual run