IBM eserver Cluster 1350, Installation And Service Manual

Page: 49 / 122

.
4. To confirm the addition, type mmlsnode -C set1.
Rules to follow when adding nodes
You must follow these rules when adding nodes to a GPFS node set:
v
A node can belong to only one node set at a time.
v The nodes being added to the node set must belong to a GPFS cluster (issue
the mmlscluster command to display available nodes).
v
The existing nodeset must meet quorum for the nodes to be added. For example,
if GPFS is currently configured on eight nodes, all of which are up and running,
the quorum value is met and the new nodes join the node set.
v Conversely, if GPFS is currently configured on eight nodes and only four are up
and running, a quorum of five does not exists and the new nodes can not join the
node set. When five of the original eight nodes are up and running, the new
nodes are added.
v
After the nodes have been added and GPFS is started on the new nodes, the
quorum value for the nodeset is adjusted accordingly. This enables new nodes to
join a running node set without causing quorum to be lost.
v
Issue the mmstartup command to start GPFS on the new nodes.
v When adding nodes to a node set using the single-node quorum algorithm, the
GPFS daemon must be stopped on all of the nodes. If after the nodes are
added, the number of nodes in the node set exceeds two, the quorum algorithm
is automatically changed to the multinode quorum algorithm.
Distributing the system image to all nodes in the cluster
Because of the way the Red Hat version 9.0 loads SCSI drivers and assigns them
to dev/sda, dev/sdb partitions, problems can result if more than one SCSI host
adapter (Adaptec or LSI SCSI controller for local drives and QLogic HBA for Triton
connection) is installed on the system. The QLogic HBA will typically be detected
first by the installation process. Follow the “Installing the storage node software” on
page 30 and modify the order of the contents of the /etc/modules.conf file.
Attempting to distribute the system image out to the nodes while a FAStT controller
is still turned on and connected might cause data damage on the first logical disk
device in FAStT subsystem. Make sure that the FAStT controllers are turned off or
that all fiber cables for the FAStT controllers are disconnected from the back of
each controller before starting the install process.
To distribute the system image to all nodes in the cluster, complete the following
steps:
1. Open an rconsole window for each node being installed so you can monitor the
installation process:
rconsole -n {node_list}
2. Run the installnode command for each node being installed:
installnode {node_list}
After the operating system is installed on the storage nodes, reconnect the fiber
cable to the FAStT controllers. Restart the storage nodes to see any configured
LANs.
Verifying the configuration
1. Start the management node and log on as user
root
.
Chapter 6. Installing the software
33