Application Performance Snapshot User Guide for Linux* OS

ID 772048
Date 10/31/2024
Public

Usage Considerations and Compatibility

This section lists important information about specific use cases for Application Performance Snapshot.

Usage Considerations

  • To profile MPI applications written in Python, include the path to your MPI library in the LD_PRELOAD command. See Analyzing Python Applications with MPI.
  • For the HTML summary report to render properly, make sure to use the latest version of your web browser.
  • To open the HTML summary report on Windows* OS from a shared drive, use a web browser other than Internet Explorer*.
  • To see metrics related to OpenMP*, you must use a version of the Intel compiler that is supported by Application Performance Snapshot. Download the latest compiler from the Intel® oneAPI Base Toolkit or the Intel® oneAPI System Bring-Up Toolkit.
  • Application Performance Snapshot uses data provided by the operating system to collect performance statistics. Application Performance Snapshot uses low precision to measure wait time or idle time caused by waiting data from the I/O subsystem. For short tasks, the I/O wait time may be zero, even if the I/O subsystem is actively used.
  • During a process run, if the system reassigns the process run to a different CPU core, the data read or written and the I/O wait time are zero.

Compatibility

Support for the use of Application Performance Snapshot may be limited in these use cases:

  • Neighborhood collective communications - Application Performance Snapshot registers the execution times only. Information about data transmission or direction of data transmission is not recorded.
  • MPI_Pcontrolregion statistics - Disk Usage and Memory Footprint metrics are not collected.

Support for the use of Application Performance Snapshot does not exist in these use cases:

  • MPI_THREAD_MULTIPLE mode
  • Intel® Trace Collector - Avoid analyzing MPI applications with both tools simultaneously.