cndaqiang Web Linux DFT

两台S9200WK服务器组Centos集群

2020-10-28
cndaqiang
RSS

配置及注意事项

  • 配置: 每个节点2*Intel(R) Xeon(R) Platinum 9242 CPU @ 2.30GHz,共96C/节点,总共192C
  • 仅有一个节点有能ipv4访问互联网的ip,两个节点均支持ipv6,因此配置安装过程尽量使用ipv6.
  • 使用有ipv4的节点做管理节点和计算节点,下面称为master节点,
    另一个节点称为client01节点
  • 环境变量仅添加到master的/etc/profile是不够的,还要添加到各个计算节点
  • 仅在master节点安装编译需要的库(如libibmad.so),需要在计算节点也安装,才能正常提交作业

重装系统Centos7

安装界面勾选IB驱动和相关的库

配置源

#ipv6源
sudo su
echo '2402:f000:1:408:8100::1 mirrors6.tuna.tsinghua.edu.cn mirrors.tuna.tsinghua.edu.cn' >> /etc/hosts
cp /etc/yum.repos.d/CentOS-Base.repo /etc/yum.repos.d/CentOS-Base.repo.bak
cat << EOF > /etc/yum.repos.d/CentOS-Base.repo
# CentOS-Base.repo
#
# The mirror system uses the connecting IP address of the client and the
# update status of each mirror to pick mirrors that are updated to and
# geographically close to the client.  You should use this for CentOS updates
# unless you are manually picking other mirrors.
#
# If the mirrorlist= does not work for you, as a fall back you can try the
# remarked out baseurl= line instead.
#
#


[base]
name=CentOS-\$releasever - Base
baseurl=https://mirrors.tuna.tsinghua.edu.cn/centos/\$releasever/os/\$basearch/
#mirrorlist=http://mirrorlist.centos.org/?release=\$releasever&arch=\$basearch&repo=os
enabled=1
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-7

#released updates
[updates]
name=CentOS-\$releasever - Updates
baseurl=https://mirrors.tuna.tsinghua.edu.cn/centos/\$releasever/updates/\$basearch/
#mirrorlist=http://mirrorlist.centos.org/?release=\$releasever&arch=\$basearch&repo=updates
enabled=1
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-7



#additional packages that may be useful
[extras]
name=CentOS-\$releasever - Extras
baseurl=https://mirrors.tuna.tsinghua.edu.cn/centos/\$releasever/extras/\$basearch/
#mirrorlist=http://mirrorlist.centos.org/?release=\$releasever&arch=\$basearch&repo=extras
enabled=1
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-7



#additional packages that extend functionality of existing packages
[centosplus]
name=CentOS-\$releasever - Plus
baseurl=https://mirrors.tuna.tsinghua.edu.cn/centos/\$releasever/centosplus/\$basearch/
#mirrorlist=http://mirrorlist.centos.org/?release=\$releasever&arch=\$basearch&repo=centosplus
gpgcheck=1
enabled=0
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-7

EOF

yum makecache
#清华源的GPG名称和centos的不同啊,这里这样解决
cp /etc/pki/rpm-gpg/RPM-GPG-KEY-CentOS-7 /etc/pki/rpm-gpg/RPM-GPG-KEY-7
#epel 源
cd /etc/pki/rpm-gpg
wget https://mirrors.tuna.tsinghua.edu.cn/epel/RPM-GPG-KEY-EPEL-7

yum -y install epel-release

sed -e 's!^metalink=!#metalink=!g' \
    -e 's!^#baseurl=!baseurl=!g' \
    -e 's!//download\.fedoraproject\.org/pub!//mirrors.tuna.tsinghua.edu.cn!g' \
    -e 's!http://mirrors\.tuna!https://mirrors.tuna!g' \
    -i /etc/yum.repos.d/epel.repo /etc/yum.repos.d/epel-testing.repo

网络

IB驱动,后续关于IB的学习会更新在Centos 配置IB网络

## 如果ib网卡不识别,或着识别后配置后无法启用,安装ib驱动
yum install -y infiniband-diags
yum install -y opensm
systemctl start opensm
systemctl enable opensm

#后期编译mvapich时需要ib库
yum install -y libibverbs
yum install -y libibverbs-devel
yum install -y libibmad-devel

master

#node7
#主机名
hostnamectl --static set-hostname master

sed -i 's/BOOTPROTO=dhcp/BOOTPROTO=static/g'  /etc/sysconfig/network-scripts/ifcfg-ib0
sed -i 's/ONBOOT=no/ONBOOT=yes/g'  /etc/sysconfig/network-scripts/ifcfg-ib0
echo IPADDR=172.16.100.7  >> /etc/sysconfig/network-scripts/ifcfg-ib0
echo PREFIX=24 >> /etc/sysconfig/network-scripts/ifcfg-ib0

client01

hostnamectl --static set-hostname client01

sed -i 's/BOOTPROTO=dhcp/BOOTPROTO=static/g'  /etc/sysconfig/network-scripts/ifcfg-ib0
sed -i 's/ONBOOT=no/ONBOOT=yes/g'  /etc/sysconfig/network-scripts/ifcfg-ib0
echo IPADDR=172.16.100.8  >> /etc/sysconfig/network-scripts/ifcfg-ib0
echo PREFIX=24 >> /etc/sysconfig/network-scripts/ifcfg-ib0

hosts

echo 172.16.100.7 master >> /etc/hosts
echo 172.16.100.8 client01 >> /etc/hosts

开机自启动脚本

vi /etc/systemd/system/rc-local.service

填入

[Unit]
Description=/etc/rc.local Compatibility
ConditionPathExists=/etc/rc.local

[Service]
Type=forking
ExecStart=/etc/rc.local start
TimeoutSec=0
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target
Alias=rc-local.service

添加执行脚本到/etc/rc.local

systemctl daemon-reload
systemctl enable rc-local
chmod +x /etc/rc.local

NIS,NFS,pbs

按照从0搭建Centos7 计算集群配置防火墙,NIS,NFS,pbs系统,使用master作为这些系统的服务器
注意:关闭SELinux 程序编译安装在/opt目录进行共享

安装编译环境

把下面的命令填入gnu-mvapich.sh执行gnu-mvapich.sh
$ROOT是安装目录,因为/opt目录是共享的,所以安装在此目录

#ROOT=$HOME/opt/gnu-mvapich
ROOT=/opt/gnu-mvapich
mkdir -p $ROOT
mkdir $ROOT/source

#创建编译脚本
cd $ROOT/source
cat << EOF > ./make.sh
if [ -e ./netlib.py ]
then
    ./setup.py --prefix=$ROOT/math --downall
else
    make -j10
    make
    make install
fi
EOF

#添加环境变量到/etc/profile或者个人bashrx或者作业脚本
MATHDIR=$ROOT/math/lib
export LD_LIBRARY_PATH=$ROOT/mvapich2-2.3.1/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/libxc-4.3.4/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/fftw-3.3.3/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/math/lib:$LD_LIBRARY_PATH
export LIBRARY_PATH=$ROOT/mvapich2-2.3.1/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/libxc-4.3.4/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/fftw-3.3.3/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/math/lib:$LIBRARY_PATH
export C_INCLUDE_PATH=$ROOT/mvapich2-2.3.1/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/libxc-4.3.4/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/fftw-3.3.3/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/math/include:$C_INCLUDE_PATH
export PATH=$ROOT/mvapich2-2.3.1/bin:$PATH
export PATH=$ROOT/fftw-3.3.3/bin:$PATH


#mvapich
cd $ROOT/source
wget https://cndaqiang.gitee.io/packages/mirrors/mvapich2/mvapich2-2.3.1.tar.gz 
tar xzvf mvapich2-2.3.1.tar.gz
cd mvapich2-2.3.1/
./configure --prefix=$ROOT/mvapich2-2.3.1 CC=gcc FC=gfortran CXX=g++
make -j40
#安装
make install


#libxc-4.3.4 for octopus-10.1
cd $ROOT/source
wget https://cndaqiang.gitee.io/packages//mirrors/libxc/libxc-4.3.4.tar.gz
tar xzvf libxc-4.3.4.tar.gz 
cd libxc-4.3.4
#--enable-shared 动态库
./configure --prefix=$ROOT/libxc-4.3.4  CC=gcc CXX=g++ FC=gfortran --enable-shared 
bash ../make.sh

#fft
cd $ROOT/source
wget https://cndaqiang.gitee.io/packages//mirrors/fftw/fftw-3.3.3.tar.gz
tar xzvf fftw-3.3.3.tar.gz
cd fftw-3.3.3/
./configure --prefix=$ROOT/fftw-3.3.3 --enable-mpi --enable-shared
bash ../make.sh



#从0编译动态库scalapack
if [ ! -d $ROOT/math ]
then
    mkdir $ROOT/math
    mkdir $ROOT/math/lib
    mkdir $ROOT/math/include
fi

cd $ROOT/source
wget https://cndaqiang.gitee.io/packages//mirrors/math/lapack-3.8.0.tar.gz
rm -rf lapack-3.8.0
tar xzvf lapack-3.8.0.tar.gz
cd lapack-3.8.0/
#修改编译参数,并默认输出动态库

#BLAS&LAPACK
echo -e "
all:libblas.so
libblas.so: \$(ALLOBJ)
\t \$(FORTRAN) -shared -Wl,-soname,\$@ -o \$@ \$(ALLOBJ)
" >> BLAS/SRC/Makefile
#TMG
echo -e "
all:libtmg.so
libtmg.so: \$(ALLOBJ)
\t \$(FORTRAN) -shared -Wl,-soname,\$@ -o \$@ \$(ALLOBJ)
" >> TESTING/MATGEN/Makefile
#LAPACK
echo -e "
all: liblapack.so
liblapack.so: \$(ALLOBJ)
\t \$(FORTRAN) -shared -Wl,-soname,\$@ -o \$@ \$(ALLOBJ)
" >> SRC/Makefile

cp make.inc.example make.inc
echo "OPTS += -fPIC" >> make.inc
echo "NOOPT += -fPIC" >> make.inc
echo -e "
MATHDIR=$ROOT/math/lib
install:all
\t cp $PWD/BLAS/SRC/libblas.so \$(MATHDIR)
\t cp $PWD/SRC/liblapack.so \$(MATHDIR)
\t cp $PWD/TESTING/MATGEN/libtmg.so \$(MATHDIR)
\t cp $PWD/*.a \$(MATHDIR)
" >> make.inc
#安装
make -j36 #必须用make,用其他参数/完成后再输make会编译测试程序,没必要
#不可以注释Makefile中
#   all: lapack_install lib blas_testing lapack_testing
#为
#   all: lapack_install lib
#   #blas_testing lapack_testing
#可以跳过测试的过程,编译更快
# 因为blas_testing依赖blas的编译
#如果非要注释,要添加依赖blas

#Scalapack
cd $ROOT/source
wget https://cndaqiang.gitee.io/packages//mirrors/math/scalapack-2.0.2.tgz
rm -rf scalapack-2.0.2
tar xzvf scalapack-2.0.2.tgz
cd scalapack-2.0.2

#echo -e "
#ALLOBJ += \$(SLASRC) \$(DLASRC)  \$(CLASRC) \$(ZLASRC)  \
#\t   \$(SCLAUX) \$(DZLAUX) \$(ALLAUX)
#all:libscalapack.so
#libscalapack.so: \$(ALLOBJ)
#\t \$(FC) -shared -Wl,-soname,\$@ -o \$@ \$(ALLOBJ)
#" >> SRC/Makefile

#使用fPIC参数,这样静态库可以转为动态库
cp SLmake.inc.example SLmake.inc
echo "FC += -fPIC" >> SLmake.inc
echo "CC += -fPIC" >> SLmake.inc
echo -e "
MATHDIR=$ROOT/math/lib
install: all
\t \$(FC) -shared -o $PWD/libscalapack.so -Wl,--whole-archive $PWD/libscalapack.a -Wl,--no-whole-archive
\t cp $PWD/libscalapack.so \$(MATHDIR)
" >> SLmake.inc
#不支持-j20,就得逐个编译,把每个库依次打包添加到libscalapack.a,没编译成功也有libscalapack.a
make  install

安装计算程序

过程略,提交作业 qsub run-qe-iop.sh

#!/bin/bash
#PBS -N QE
#PBS -l nodes=2:ppn=96
#PBS -q regular
#PBS -j oe
#PBS -l walltime=08:00:00

ROOT=/opt/gnu-mvapich
MATHDIR=$ROOT/math/lib
export LD_LIBRARY_PATH=$ROOT/mvapich2-2.3.1/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/libxc-4.3.4/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/fftw-3.3.3/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$ROOT/math/lib:$LD_LIBRARY_PATH
export LIBRARY_PATH=$ROOT/mvapich2-2.3.1/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/libxc-4.3.4/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/fftw-3.3.3/lib:$LIBRARY_PATH
export LIBRARY_PATH=$ROOT/math/lib:$LIBRARY_PATH
export C_INCLUDE_PATH=$ROOT/mvapich2-2.3.1/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/libxc-4.3.4/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/fftw-3.3.3/include:$C_INCLUDE_PATH
export C_INCLUDE_PATH=$ROOT/math/include:$C_INCLUDE_PATH
export PATH=$ROOT/mvapich2-2.3.1/bin:$PATH
export PATH=$ROOT/fftw-3.3.3/bin:$PATH


EXEC=~/code/tdpw2.0/q-e-qe-6.4.1-devel-1.0/bin/tdpw.x
#EXEC=pw.x
cd $PBS_O_WORKDIR

NP=`cat $PBS_NODEFILE | wc -l`

echo Job starts at `date`

#mpirun -np $NP -machinefile $PBS_NODEFILE
mpirun  $EXE < input.in | tee  result
echo Job ends at `date`

192核一起运行,隔音墙都挡不住的噪声

2020-11-06 方案二使用slurm

Centos7集群上搭建slurm作业管理系统

2020-11-07 挂载算盘的NFS存储,使用master的NIS服务

仍然采用slurm,但是挂载算盘的存储,后期将master转为slurm管理节点,将算盘的几个计算节点转移过来

master

配置ib网卡,没想到居然兼容

vi /etc/sysconfig/network-scripts/ifcfg-ib0
/etc/init.d/network restart

和之前的两个机器只连相比,速度从56Gb/sec (4X FDR)降低到40 Gb/sec (4X QDR)

[root@master cndaqiang]# ibstatus
Infiniband device 'mlx5_0' port 1 status:
	default gid:	 fe80:0000:0000:0000:b859:9f03:00d0:135e
	base lid:	 0x2
	sm lid:		 0x2
	state:		 4: ACTIVE
	phys state:	 5: LinkUp
	rate:		 40 Gb/sec (4X QDR)
	link_layer:	 InfiniBand

hosts

10.10.10.107 master
10.10.10.108 client01
10.10.10.100 mu01 #存储节点

NIS服务

也许因为以前服务器版本太老老,失败,继续使用master做NIS服务

NFS服务

master节点不再担任任何的存储服务 因为服务器的NFS版本过老,直接挂载会出现nobody,不适合/etc/fstab方式挂载 理想的方式

mkdir /public
mount -t nfs -o vers=3 ibmu01:/home /public
echo "mount -t nfs -o vers=3 ibmu01:/home /public" >> /etc/rc.local
mkdir -p /public/master/opt
mkdir -p /public/master/home
ln -s /public/master/opt /opt

添加到/etc/fstab

#ibmu01:/home /public nfs      defaults        0 0

以后新建用户使用

adduser --home /public/master/home/cndaqiang2 cndaqiang2

修改已有账户主目录

usermod -d /public/master/home/cndaqiang -u $(id -u cndaqiang) cndaqiang
mkdir /public/master/home/cndaqiang
chown cndaqiang:cndaqiang /public/master/home/cndaqiang
su cndaqiang
#复制之前的数据
for i in $(ls -a /home/cndaqiang/ | grep -v '\.\.' | grep -v '\.$'); do echo $i ; cp -r /home/cndaqiang/$i $HOME; done

更新用户信息

make -C /var/yp

slurm

Centos7集群上搭建slurm作业管理系统


本文首发于我的博客@cndaqiang.
本博客所有文章除特别声明外,均采用 CC BY-SA 4.0 协议 ,转载请注明出处!


类似文章


评论


广告


广告
访客数据