11. Run-time operation and tuning MPI applications

Once your MPI application is running, there are many different components that can be tuned to affect behavior at run time.

• 11.1. Environment variables set for MPI applications
• 11.2. Networking support
• 11.3. Accelerator support
• 11.4. Collective operations
• 11.5. Running multi-threaded MPI applications
• 11.6. Dynamically loading libmpi at runtime
• 11.7. Calling fork(), system(), or popen() in MPI processes
• 11.8. Fault tolerance
• 11.9. Large Clusters
• 11.10. Processor and memory affinity
• 11.11. MPI IO
• 11.12. Benchmarking Open MPI applications
• 11.13. Building heterogeneous MPI applications