“Steps of adding a new grid node”版本间的差异

2014年9月5日 (五) 02:03的版本

using a backup image

1. restore image when booting from USB

1. post installation settings
   1. do a full system update (yum update -y)
   2. reset root password
   3. change /etc/fstab, make directory /work{,0,1,2}
   4. install sge under /opt/sge
      1. set cell name to 'redhat' when run /opt/sge/install_execd
      2. rm -f /etc/profile.d/sge.{,c}sh && cp /opt/sge/redhat/common/settings.{,c}sh /etc/profile.d
   5. yum install -y ntp && ntpdate pool.ntp.org && hwclock -w
   6. ln -s /nfs/disk/perm/sys/env/cslt.sh /etc/profile.d (grid without GPU) or ln -s /nfs/disk/perm/sys/env/cslt_cuda.sh /etc/profile.d (grid with GPU)

generic steps

1. Install Linux box grid-X,

 * Ensure the following are there: perl, python, g++ for c/c++/fortran, Tcl.
 * mount nfs service from grid-0 to  /nfs/disk
 * define the node name as grid-X and update /etc/hosts for the nodes to reflect the new node.
 * copy /etc/bash.bashrc from grid-0 to the new node

2. Install NIS service, propagate the NIS account from grid-0

3. Install SGE, accepted by grid-0

4. Install additional tools

 * matlab client
 * ATALAS and CLAPACK

5. Done

some tips:

1. statistic IPs are used right now. For that configure /etc/conf.d/netcfg. Check /etc/network.d/examples to create a profile. systemctl enable netcfg is needed to enable the service. DHCP should be disabled, i.e., systemctl disable dhcp@eth0.

2. in the old nodes, grid-0 and gird-1, the netcfg is enabled by replacing network by net-profiles in DAEMON of /etc/rc.conf.