Tutorial

Using MPI Tuner for Intel® MPI Library on Linux* OS

Download PDF
ID 768652
Date 2/19/2016
Public
Document Table of Contents
Document Table of Contents x
Tutorial: Using MPI Tuner for Intel® MPI Library
Tutorial: Using MPI Tuner for Intel® MPI Library x
Legal Information Overview Getting Started Task 1: Minimize Tuning Time in Cluster- Specific Mode Task 2: Include Missing Values in the Default Parameters Grid during Cluster Tuning Task 3: Application-Specific Tuning Troubleshooting
Getting Started x
Accessing MPI Tuner MPI Tuner Modes
MPI Tuner Modes x
Cluster-Specific Tuning Commands Application-Specific Tuning Commands
Task 1: Minimize Tuning Time in Cluster- Specific Mode x
Host Range Numbers of Ranks per Host Fabric Usage (I_MPI_FABRICS) Message Sizes Most Common MPI Functions
Task 3: Application-Specific Tuning x
Commands for Fast Application-Specific Tuning Commands for Topology Awareness Rank Placement Optimization
Tutorial: Using MPI Tuner for Intel® MPI Library

MPI Tuner Modes

The MPI tuner utility can operate in four modes:

  • Cluster-specific. Evaluates a given cluster environment using either Intel® MPI Benchmarks, or a user- provided benchmarking program to find the most suitable configuration of Intel® MPI Library. This mode is used by default.
  • Application-specific. Evaluates performance of a given MPI application to find the best configuration of Intel® MPI Library for a particular application.
  • Fast application-specific. In contrast to the previous mode, instead of launching a real application this mode performs micro-kernel tests during the tuning process. The set of tests is based on the Intel® MPI Library statistics collected from an application run. This approach is not very accurate but can speedup applications with non-typical patterns (non-typical rank placement per host or in a communicator, unusual messages sizes, and so on). This mode supports only regular collective operations that have related environment variables in the I_MPI_ADJUST family (excluding the KN_RADIX subset and I_MPI_ADJUST_REDUCE_SEGMENT). See the Intel® MPI Library Developer Reference for details.
  • Topology-aware rank placement optimization. Evaluates cluster characteristics and rank-to-rank data transfers in an MPI application to identify the optimal rank placement. This mode is recommended for applications where the communication pattern has some kind of local groups, such as, stencils, collective operations on subsets of ranks, neighborhood operations, and so on.
Parent topic: Getting Started
Level Two Title