本文讨论了一个方法，用于优化与英特尔®MPI 库连接的编程应用的消息传递接口（MPI）集合性能。 英特尔 MPI 库添加了 I_MPI _ADJUST 环境变量家族，允许对与其每个群体操作相关的算法进行显式选择。 使用英特尔 MPI 库时，MPI 集合优化的方法通过添加进程固定和互联架构控制，从而测量英特尔® MPI 基准测试的消息传递延迟进行了证明。 针对集合的英特尔 MPI 基准测试分析是在英特尔® 至强融核™ 协处理器的本机模式下运行的，但这种选择最佳集合算法的模式适用于其他英特尔® 微架构。
NOTE: Defects and feature requests described below represent specific issues with specific test cases. It is difficult to succinctly describe an issue and how it impacted the specific test case. Some of the issues listed may impact multiple architectures, operating systems, and/or languages. If you have any questions about the issues discussed in this report, please post on the user forums, http://software.intel.com/en-us/forums or submit an issue to Intel® Premier Support, https://premier.intel.com.
If you have a previous version of the Intel® MPI Library for Linux* OS installed, you do not need to uninstall it before installing the latest version.
Extract the l_mpi[-rt]_p_<version>.<package_num>.tar.gz package by using following command:
tar –xvzf l_mpi[-rt]_p_<version>.<package_num>.tar.gz
This command creates the subdirectory l_mpi[-rt]_p_<version>.<package_num>.
Use the -t or -trace option to link the resulting executable files against the Intel® Trace Collector library. This has the same effect as when -profile=vt is used as an argument to mpiicc or another compiler script.
$ mpiicc -trace test.c -o testc
To use this option, you need to:
To limit the number of nodes on which tuning will be performed, use the --host-range min:max option. For example, the following command will restrict running on 8 to 16 nodes only:
$ mpitune --host-range 8:16
cannot execute binary file
Wrong binary executable file format or architecture.
This error occurs when you run a binary executable file build for x86_64 architecture on a k1om node (for example, on an Intel® Xeon Phi™ coprocessor). In such cases, hang of the MPI application is possible.
You can select a profile library through the -profile option of the Intel® MPI Library compiler drivers. The profile files are located in the <installdir>/<arch>/etc directory. The Intel® MPI Library comes with several predefined profiles for the Intel® Trace Collector:
<installdir>/etc/vt.conf - regular Intel® Trace Collector library
Checkpoint images are large; thus, Intel® MPI Library only keeps the last useful checkpoint by default. The following is an example to keep <N> previous checkpoints. The flag is -ckpoint-preserve <N>. The default value of -ckpoint-preserve is 1 (keep only the last checkpoint).
The Intel® MPI Library provides an automatic tuning utility to help you select optimal values for many environment variables that can be used to influence program behavior and performance at run time.