1.安装GPFS包

#rpm –ivh *.rpm
(gpfs.base-3.1.0-1.x86_64,gpfs.docs-3.1.0-1.noarch,gpfs.gpl-3.1.0-1.noarch,gpfs.msg.en_US-3.1.0-1.noarch)
2. 安装GPFS升级包

rpm –Uvh *.rpm
(gpfs.base-3.1.0-8.x86_64,gpfs.docs-3.1.0-8.noarch,gpfs.gpl-3.1.0-8.noarch,gpfs.msg.en_US-3.1.0-8.noarch)

3. 编译

#cd /usr/lpp/mmfs/src/config
#cp site.mcr.proto site.mcr 
#vi site.mcr
参考 README 修改site.mcr
#cd /usr/lpp/mmfs/src
#export SHARKCLONEROOT=/usr/lpp/mmfs/src
#make Autoconfig
#make InstallImages
#cd /usr/lpp/mmfs/bin

/* Copy bin files to all the other nodes */

4. 系统优化：

mmchconfig pagepool=1000M （4M－1/2内存总量）
mmchconfig maxFilesToCache ＝1700（default,1-100000）
maxStatCache (4 × maxFilesToCache )
Aggregate network interfaces 
Swap space (2 x total of mem)

updatedb considerations (skip gpfs) important

Suggested starting points are: 
mmchconfig prefetchThreads=18
mmchconfig worker1Threads=24 

Verify /proc/sys/net/ipv4/tcp_window_scaling is enabled. It should be by default. 

Tune the TCP window settings by adding these lines to the /etc/sysctl.conf file:
# increase Linux TCP buffer limits 
net.core.rmem_max = 8388608 
net.core.wmem_max = 8388608 
# increase default and maximum Linux TCP buffer sizes 
net.ipv4.tcp_rmem = 4096 262144 8388608 
net.ipv4.tcp_wmem = 4096 262144 8388608 
# increase max backlog to avoid dropped packets 
net.core.netdev_max_backlog=2500 

A. Issue the sysctl -p /etc/sysctl.conf command to set the kernel settings. 
B. Issue the mmstartup -a command to restart GPFS 


5 .创建GPFS集群

#/usr/lpp/mmfs/bin/mmcrcluster -t lc -p io1 -s io2 -n
/tmp/gpfs.nodes -r /usr/bin/ssh -R /usr/bin/scp
List the GPFS cluster details using:
#/usr/lpp/mmfs/bin/mmlscluster

6. 启动 GPFS
mmstartup -a

7. 磁盘定义
cat /tmp/gpfs.disks
/dev/sda:io1::dataAndMetadata:-1
/dev/sdb:io2::dataAndMetadata:-1
....
#cp /tmp/gpfs.disks /tmp/gpfs.disks.bak
#/usr/lpp/mmfs/bin/mmcrnsd -F /dump/gpfs.disks -v yes
#mmlsnsd

8.创建GPFS文件系统
# mmcrfs /gpfs1 gpfs1 -F /tmp/gpfs.disks1 -A yes

9.增加GPFS客户端
#mmaddnode -N node1
#mmaddnode -N node2