---
title: mpi4py安裝使用說明 | zh
tags: Guide, TWNIA3, TW
GA:
---
<p style="text-align:right;">更新日期：Feb. 29, 2024</p>
# mpi4py安裝使用
:::warning
:warning: 執行環境須與編譯安裝的環境設定一致，勿混用！！！
* 使用conda安裝者，請以conda環境執行。
* 使用pip方式安裝者，請以安裝時的環境設定執行。
:::
[TOC]
## conda 安裝方式與執行
透過conda安裝是最簡單快速，缺點是mpi4py、MPI library版本受限於conda版本，無法隨意選擇版本組合。
1. 登入系統載入conda模組，並建立欲執行的環境
:::spoiler 點此查看範例，以下同。
```
[demo_user@lgn301 ~]$ ml pkg/Anaconda3 # 載入conda模組
(base) [demo_user@lgn301 ~]$ conda create -n demo # 建立conda demo環境
Collecting package metadata (current_repodata.json): done
Solving environment: done
...（略）...
## Package Plan ##
environment location: /home/demo_user/.conda/envs/demo
Proceed ([y]/n)? y
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
# $ conda activate demo
#
# To deactivate an active environment, use
#
# $ conda deactivate
```
:::
2. 安裝MPI library
可省略此一步驟，直接安裝mpi4py，若之前未安裝過MPI library，預設是安裝mpich。
:::spoiler 範例
```
[demo_user@lgn301 ~]$ conda activate demo #啟用 demo 的conda環境設定
(demo) [demo_user@lgn301 ~]$ conda install openmpi # 安裝OpenMPI
Collecting package metadata (current_repodata.json): done
Solving environment: done
...（略）...
## Package Plan ##
environment location: /home/demo_user/.conda/envs/demo
added / updated specs:
- openmpi
The following packages will be downloaded:
package | build
---------------------------|-----------------
_openmp_mutex-5.1 | 1_gnu 21 KB
libgcc-ng-11.2.0 | h1234567_1 5.3 MB
libgfortran-ng-7.5.0 | ha8ba4b0_17 22 KB
libgfortran4-7.5.0 | ha8ba4b0_17 995 KB
libgomp-11.2.0 | h1234567_1 474 KB
libstdcxx-ng-11.2.0 | h1234567_1 4.7 MB
mpi-1.0 | openmpi 7 KB
openmpi-4.0.2 | hb1b8bf9_1 3.5 MB
zlib-1.2.13 | h5eee18b_0 103 KB
------------------------------------------------------------
Total: 15.1 MB
The following NEW packages will be INSTALLED:
_libgcc_mutex pkgs/main/linux-64::_libgcc_mutex-0.1-main
_openmp_mutex pkgs/main/linux-64::_openmp_mutex-5.1-1_gnu
libgcc-ng pkgs/main/linux-64::libgcc-ng-11.2.0-h1234567_1
libgfortran-ng pkgs/main/linux-64::libgfortran-ng-7.5.0-ha8ba4b0_17
libgfortran4 pkgs/main/linux-64::libgfortran4-7.5.0-ha8ba4b0_17
libgomp pkgs/main/linux-64::libgomp-11.2.0-h1234567_1
libstdcxx-ng pkgs/main/linux-64::libstdcxx-ng-11.2.0-h1234567_1
mpi pkgs/main/linux-64::mpi-1.0-openmpi
openmpi pkgs/main/linux-64::openmpi-4.0.2-hb1b8bf9_1
zlib pkgs/main/linux-64::zlib-1.2.13-h5eee18b_0
Proceed ([y]/n)? y
Downloading and Extracting Packages
_openmp_mutex-5.1 | 21 KB | ############################### | 100%
mpi-1.0 | 7 KB | ############################### | 100%
libgcc-ng-11.2.0 | 5.3 MB | ############################### | 100%
zlib-1.2.13 | 103 KB | ############################### | 100%
libstdcxx-ng-11.2.0 | 4.7 MB | ############################### | 100%
openmpi-4.0.2 | 3.5 MB | ############################### | 100%
libgomp-11.2.0 | 474 KB | ############################### | 100%
libgfortran-ng-7.5.0 | 22 KB | ############################### | 100%
libgfortran4-7.5.0 | 995 KB | ############################### | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
```
:::
3. 安裝mpi4py
:::spoiler 範例
```
(demo) [demo_user@lgn301 ~]$ conda install mpi4py # 安裝mpi4py
Collecting package metadata (current_repodata.json): done
Solving environment: done
...（略））...
## Package Plan ##
environment location: /home/demo_user/.conda/envs/demo
added / updated specs:
- mpi4py
The following packages will be downloaded:
package | build
---------------------------|-----------------
bzip2-1.0.8 | h5eee18b_5 262 KB
ca-certificates-2023.12.12 | h06a4308_0 126 KB
ld_impl_linux-64-2.38 | h1181459_1 654 KB
libffi-3.4.4 | h6a678d5_0 142 KB
libuuid-1.41.5 | h5eee18b_0 27 KB
mpi4py-3.1.4 | py311h3e5f7c9_0 541 KB
ncurses-6.4 | h6a678d5_0 914 KB
openssl-3.0.13 | h7f8727e_0 5.2 MB
pip-23.3.1 | py311h06a4308_0 3.3 MB
python-3.11.8 | h955ad1f_0 32.9 MB
readline-8.2 | h5eee18b_0 357 KB
setuptools-68.2.2 | py311h06a4308_0 1.2 MB
sqlite-3.41.2 | h5eee18b_0 1.2 MB
tk-8.6.12 | h1ccaba5_0 3.0 MB
tzdata-2024a | h04d1e81_0 116 KB
wheel-0.41.2 | py311h06a4308_0 145 KB
xz-5.4.6 | h5eee18b_0 651 KB
------------------------------------------------------------
Total: 50.7 MB
The following NEW packages will be INSTALLED:
bzip2 pkgs/main/linux-64::bzip2-1.0.8-h5eee18b_5
ca-certificates pkgs/main/linux-64::ca-certificates-2023.12.12-h06a4308_0
ld_impl_linux-64 pkgs/main/linux-64::ld_impl_linux-64-2.38-h1181459_1
libffi pkgs/main/linux-64::libffi-3.4.4-h6a678d5_0
libuuid pkgs/main/linux-64::libuuid-1.41.5-h5eee18b_0
mpi4py pkgs/main/linux-64::mpi4py-3.1.4-py311h3e5f7c9_0
ncurses pkgs/main/linux-64::ncurses-6.4-h6a678d5_0
openssl pkgs/main/linux-64::openssl-3.0.13-h7f8727e_0
pip pkgs/main/linux-64::pip-23.3.1-py311h06a4308_0
python pkgs/main/linux-64::python-3.11.8-h955ad1f_0
readline pkgs/main/linux-64::readline-8.2-h5eee18b_0
setuptools pkgs/main/linux-64::setuptools-68.2.2-py311h06a4308_0
sqlite pkgs/main/linux-64::sqlite-3.41.2-h5eee18b_0
tk pkgs/main/linux-64::tk-8.6.12-h1ccaba5_0
tzdata pkgs/main/noarch::tzdata-2024a-h04d1e81_0
wheel pkgs/main/linux-64::wheel-0.41.2-py311h06a4308_0
xz pkgs/main/linux-64::xz-5.4.6-h5eee18b_0
Proceed ([y]/n)? y
Downloading and Extracting Packages
bzip2-1.0.8 | 262 KB | ############################### | 100%
python-3.11.8 | 32.9 MB | ############################### | 100%
openssl-3.0.13 | 5.2 MB | ############################### | 100%
sqlite-3.41.2 | 1.2 MB | ############################### | 100%
pip-23.3.1 | 3.3 MB | ############################### | 100%
wheel-0.41.2 | 145 KB | ############################### | 100%
ncurses-6.4 | 914 KB | ############################### | 100%
readline-8.2 | 357 KB | ############################### | 100%
ld_impl_linux-64-2.3 | 654 KB | ############################### | 100%
libuuid-1.41.5 | 27 KB | ############################### | 100%
xz-5.4.6 | 651 KB | ############################### | 100%
tzdata-2024a | 116 KB | ############################### | 100%
setuptools-68.2.2 | 1.2 MB | ############################### | 100%
mpi4py-3.1.4 | 541 KB | ############################### | 100%
tk-8.6.12 | 3.0 MB | ############################### | 100%
ca-certificates-2023 | 126 KB | ############################### | 100%
libffi-3.4.4 | 142 KB | ############################### | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
```
:::
4. 測試
測試程式參考自[Github](https://github.com/jbornschein/mpi4py-examples)
:::spoiler 範例
```
(demo) [demo_user@lgn301 ~]$ cd test
(demo) [demo_user@lgn301 test]$ cat hello.py
#!/usr/bin/env python
from __future__ import print_function
from mpi4py import MPI
comm = MPI.COMM_WORLD
print("Hello! I'm rank %d from %d running in total..." % (comm.rank, comm.size))
comm.Barrier() # wait for everybody to synchronize _here_
(demo) [demo_user@lgn301 test]$ which mpirun
~/.conda/envs/demo/bin/mpirun
(demo) [demo_user@lgn301 test]$ which python3
~/.conda/envs/demo/bin/python3
(demo) [demo_user@lgn301 test]$ mpirun -n 4 python3 hello.py
Hello! I'm rank 0 from 4 running in total...
Hello! I'm rank 1 from 4 running in total...
Hello! I'm rank 2 from 4 running in total...
Hello! I'm rank 3 from 4 running in total...
```
:::
5. 撰寫job script
:::spoiler 範例
```
(demo) [demo_user@lgn301 test]$ cat mpi4py.sh
#!/bin/bash
#SBATCH -A ABC777777 # Account name/project number
#SBATCH -J mpi4py_test # Job name
#SBATCH -p ctest
#SBATCH -n 4 # Number of MPI tasks (i.e. processes)
#SBATCH -c 1 # Number of cores per MPI task
#SBATCH -o %j.out # Path to the standard output and error files relative to the working directory
#SBATCH -e %j.err # Path to the standard output and error files relative to the working directory
ml purge
ml pkg/Anaconda3
conda activate demo
mpirun -n 4 python hello.py
```
:::info
:bulb:若是安裝mpich 請以mpiexec執行
:::
6. 提交job script
(demo) [demo_user@lgn301 test]$ sbatch mpi4py.sh
Submitted batch job 9004003
7. 查看結果
:::spoiler 點此查看
```
[demo_user@lgn301 test]$ cat 9004003.out
Hello! I'm rank 1 from 4 running in total...
Hello! I'm rank 2 from 4 running in total...
Hello! I'm rank 3 from 4 running in total...
Hello! I'm rank 0 from 4 running in total...
```
:::
## pip 安裝方式與執行
透過pip3安裝可自由選擇編譯器、MPI library、Python版本組合，尤其是節點間若有安裝高速網路介面者，可藉由支援的特殊協定達成快速交換資料訊息。
1. 登入系統，載入欲使用的編譯器、MPI library與Python
本例係以Intel oneAPI compiler搭配 Intel oneAPI MPI為示範，亦可使用gcc、clang...等編譯器搭配其他MPI，如OpenMPI、MPICH、MVAPICH2...等
```
[demo_user@lgn302 ~]$ ml intel intelmpi Python
[demo_user@lgn302 ~]$ ml
Currently Loaded Modules:
1) intel/2024 2) intelmpi/2021.11 3) Python/3.12.2
```
2. 安裝mpi4py至個人目錄下
```
[demo_user@lgn302 ~]$ pip3 list
Package Version
------- -------
pip 24.0
[demo_user@lgn302 ~]$ pip3 install mpi4py --user ## 安裝mpi4py至個人目錄
Collecting mpi4py
Using cached mpi4py-3.1.5-cp312-cp312-linux_x86_64.whl
Installing collected packages: mpi4py
Successfully installed mpi4py-3.1.5
[demo_user@lgn302 ~]$ pip3 list
Package Version
------- -------
mpi4py 3.1.5
pip 24.0
```
3. 撰寫job script
測試程式參考自[Github](https://github.com/jbornschein/mpi4py-examples)
:::spoiler 點此查看範例
```
[demo_user@lgn302 ~]$ cd test
[demo_user@lgn302 test]$ cat hello.py
#!/usr/bin/env python
from __future__ import print_function
from mpi4py import MPI
comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
node_name = MPI.Get_processor_name() # get the name of the node
print("Hello world! I'm rank %d from %d running in total at %s." % (comm.rank, comm.size, node_name))
#print("Hello! I'm rank %d from %d running in total..." % (comm.rank, comm.size))
comm.Barrier() # wait for everybody to synchronize _here_
```
```
[demo_user@lgn302 test]$ cat impi4py.sh
#!/bin/bash
#SBATCH -A ABC123456 # Account name/project number
#SBATCH -J mpi4py_test # Job name
#SBATCH -p hetero
#SBATCH -n 8 # Number of MPI tasks (i.e. processes)
#SBATCH -c 1 # Number of cores per MPI task
#SBATCH -N 2 # Maximum number of nodes to be allocated
#SBATCH -o %j.out # Path to the standard output and error files relative to the working directory
#SBATCH -e %j.err # Path to the standard output and error files relative to the working directory
ml purge
ml intel/2024 intelmpi/2021.11 Python/3.12.2
mpirun python3 hello.py
```
:::
4. 提交計算工作
[demo_user@lgn302 test]$ sbatch impi4py.sh
Submitted batch job 9006770
5. 查看結果
:::spoiler 點此查看結果
```
[demo_user@lgn302 test]$ cat 9006770.out
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
MPI startup(): Warning: I_MPI_PMI_LIBRARY will be ignored since the hydra process manager was found
Hello world! I'm rank 0 from 8 running in total at cpn3001.cluster.
Hello world! I'm rank 1 from 8 running in total at cpn3001.cluster.
Hello world! I'm rank 4 from 8 running in total at cpn3002.cluster.
Hello world! I'm rank 2 from 8 running in total at cpn3001.cluster.
Hello world! I'm rank 5 from 8 running in total at cpn3002.cluster.
Hello world! I'm rank 3 from 8 running in total at cpn3001.cluster.
Hello world! I'm rank 6 from 8 running in total at cpn3002.cluster.
Hello world! I'm rank 7 from 8 running in total at cpn3002.cluster.
```
:::
:::warning
:warning: 執行環境須與編譯安裝的環境設定一致，勿混用！！！
* 使用conda安裝者，請以cond環境執行。
* 使用pip方式安裝者，請以安裝時的環境設定執行。
* 須跨節點執行者，建議以pip3方式安裝。
:::