Analyzing an MPI Application

Use Intel® Trace Analyzer to conduct a performance analysis and tune MPI communications between processes. Then switch to the Intel® VTune™ Amplifier XE to carry out a new round of analysis and identify hotspots on specific processes.

This tutorial uses the poisson sample code as well as the sample trace files poisson_sendrecv.single.stf and poisson_icomm.single.stf to demonstrate the interoperability of the Intel Trace Analyzer and Intel VTune Amplifier XE that enables you to analyze and further tune the application.

Step 1: Optimize MPI communications

  • Detect serialization in Function Profile and Message Profile.

  • Find imbalance on the inter-process level by comparing the original application with the idealized one.

  • Remove serialization.

Step 2: Improve intra-process performance

  • Identify hotspots on the intra-process level.

  • Interpret the collected data to find MPI processes in the workload and find possible ways to resolve the issue.

For more complete information about compiler optimizations, see our Optimization Notice.