Intel® MPI Library Developer Guide for Linux* OS

Download PDF
ID 768728
Date 7/13/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents
Document Table of Contents x
Intel® MPI Library Developer Guide for Linux* OS
Intel® MPI Library Developer Guide for Linux* OS x
Introduction Installation and Prerequisites Compiling and Linking Running Applications Debugging Applications Analysis and Tuning Troubleshooting Additional Supported Features Examples Notices and Disclaimers
Introduction x
Introducing Intel® MPI Library Conventions and Symbols Related Information
Installation and Prerequisites x
Installation Prerequisite Steps
Compiling and Linking x
Compiling an MPI Program Compilers Support ILP64 Support
Running Applications x
Running Intel® MPI Library in Containers Selecting a Library Configuration Running an MPI Program Running an MPI/OpenMP* Program MPMD Launch Mode Fabrics Control Job Schedulers Support Controlling Process Placement Java* MPI Applications Support
Running Intel® MPI Library in Containers x
Build a Singularity* Container for an MPI Application Run the Application with a Container
Fabrics Control x
Selecting Fabrics Libfabric* Support OFI* Providers Support
Debugging Applications x
Debugging Using -gtool for Debugging
Analysis and Tuning x
Displaying MPI Debug Information Tracing Applications Interoperability with Other Tools Through -gtool MPI Tuning Autotuner mpitune_fast mpitune
Troubleshooting x
Error Message: Bad Termination Error Message: No such file or Directory Error Message: Permission Denied Error Message: Fatal Error Error Message: Bad File Descriptor Error Message: Too Many Open Files Problem: High Memory Consumption Readings Problem: MPI Application Hangs Problem: Password Required Problem: Cannot Execute Binary File Problem: MPI limitation for Docker*
Additional Supported Features x
Asynchronous Progress Control Multiple Endpoints Support
Multiple Endpoints Support x
MPI_THREAD_SPLIT Programming Model Threading Runtimes Support Program Examples Code Change Guide
Examples x
async_progress_sample.c thread_split.cpp thread_split_omp_for.c thread_split_omp_task.c thread_split_pthreads.c
Intel® MPI Library Developer Guide for Linux* OS

MPI Tuning

Intel® MPI Library provides the following tuning utilities:

Autotuner

Autotuner is the recommended utility for the application-specific tuning. If an application is spending significant time in MPI collective operations, autotuning might improve its performance. Autotuner is easy-to-use, and its overhead is close to zero.

The autotuning utility's tuning scope is I_MPI_ADJUST_<opname> family of environment variables, which are MPI collective operation algorithms. Autotuner limits tuning to the current cluster configuration (fabric, number of ranks, number of ranks per node). It works while an application is running, so performance could be potentially improved just by enabling the autotuner. It is also possible to generate new tuning file with MPI collective operations adjusted to application needs, and this file can be further passed to the I_MPI_TUNING_BIN variable.

mpitune_fast

mpitune_fast is the recommended easy-to-use utility for the cluster-wide tuning. It uses the autotuner internally, so its search space is also collective operation algorithms. mpitune_fast iteratively launches IMB with options provided (e.g., scale of tuning and collective operations to tune) and generates a file with tuning parameters for cluster configuration. This file could be provided to the Intel MPI Library with the I_MPI_TUNING_BIN environment variable. mpitune_fast supports Slurm* and LSF* workloads managers and should automatically detect job-allocated hosts to use. mpitune_fast can also perform validation of new tuning files and generate CSV files with performance results, so you do not have to validate tuning manually.

mpitune

mpitune is useful If the search space of the autotuner is not sufficient for your needs. mpitune iteratively launches a benchmarking application with different configurations to measure performance and stores the results of each launch. Based on these results, the tuner generates optimal values for parameters that are being tuned. mpitune has an ability to search for optimal values of variables other than I_MPI_ADJUST_<opname>, and it could be used for application-specific and cluster-wide tuning. For example, it could tune parameters (like radix) of collective operations.

Key difference between mpitune and mpitune_fast is that mpitune run application N times, where N is the number of possible variable values, while mpitune_fast finds the optimal I_MPI_ADJUST_<opname> after running IMB for that collective operation only once.

Differences between the tuning utilities:

Parameter Autotuner mpitune_fast mpitune
Low tuning overhead + + -
Ease of use + + -
Application tuning + - +
Microbenchmark tuning + + +
Tuning beyond collective operations - - +

Parent topic: Analysis and Tuning
Level Two Title