NOTE:
The LUN underlying the basic disk should be presented to only one node of the cluster
using selective storage presentation or SAN zoning, or having only one node online at all times
until the physical resource for the basic disk is established.
In preparing for the cluster installation:
•
All shared disks, including the Quorum disk, must be accessible from all nodes. When testing
connectivity between the nodes and the LUN, only one node should be given access to the
LUN at a time.
•
All shared disks must be configured as basic (not dynamic).
•
All partitions on the disks must be formatted as NTFS.
Network planning
Clusters require more sophisticated networking arrangements than a stand alone storage system.
A Windows NT domain or Active Directory domain must be in place to contain the cluster names,
virtual server names, and user and group information. A cluster cannot be deployed into a non
domain environment.
All cluster deployments have at least six network addresses and four network names:
•
The cluster name (Unique NETBIOS Name) and IP address
•
Node A's name and IP address
•
Node B's name and IP address
•
At least one virtual server name and IP address for virtual server A
•
Cluster Interconnect static IP addresses for Node A and Node B
In multi-node deployments, additional network addresses are required. For each additional node,
three static IP addresses are required.
Virtual names and addresses are the only identification used by clients on the network. Because
the names and addresses are virtual, their ownership can transition from one node to the other
during a failover, preserving access to the resources in the cluster group.
A cluster uses at least two network connections on each node:
•
The private cluster interconnect or “heartbeat” crossover cable connects to one of the network
ports on each cluster node. In more than two node deployments, a private VLAN on a switch
or hub is required for the cluster interconnect.
•
The public client network subnet connects to the remaining network ports on each cluster node.
The cluster node names and virtual server names have IP addresses residing on these subnets.
NOTE:
If the share is to remain available during a failover, each cluster node must be connected
to the same network subnet. It is impossible for a cluster node to serve the data to a network to
which it is not connected.
Protocol planning
Not all file sharing protocols can take advantage of clustering. If a protocol does not support
clustering, it will not have a cluster resource and will not failover with any cluster group. In the
case of a failover, a client cannot use the virtual name or virtual IP address to access the share
since the protocol cannot failover with the cluster group. The client must wait until the initial node
is brought back online to access the share.
HP recommends placing cluster aware and non cluster aware protocols on different file shares.
Cluster planning
63
