Networking Capacity; Network Considerations; Network Configuration; Network Clients - Compaq ProLiant Clusters HA/F100 Administrator's Manual

Networking Capacity

The final capacity planning section addresses networking. The cluster nodes
must have enough network capacity to handle requests from the client
machines and must gracefully handle failover/failback events.
Make sure both nodes can handle the maximum number of clients that can
attach to the cluster. If Node 1 encounters a failure and its applications and
services fail over to Node 2, then Node 2 needs to handle access from its own
network clients as well as those that normally connect to the failed node
(Node 1).
Note the effect of failover on network I/O bandwidth. When the cluster
encounters a server failover event, only one node is responding to network I/O
requests. Be sure the surviving node's network speed and protocol will
sufficiently handle the maximum number of network I/Os when the cluster is
running in a degraded state.

Network Considerations

This section addresses clustering items that affect the corporate LAN. The
Microsoft clustering software has specific requirements regarding which
protocol can be used and how IP address and network name resolution occurs.
Additionally, consider how network clients will interact within a clustering
environment. Some client-side applications may need modification to receive
the maximum availability benefits of operating a cluster.

Network Configuration

Network Protocols
TCP/IP and NBT (NetBIOS over TCP/IP) are the only transport protocols that
are supported in an Microsoft clustering software failover environment. Other
protocols, such as NetBEUI, IPX/SPX (Novell), NB/IPX, or DLC (IBM) may
be used, but they cannot take advantage of the failover features of the
Microsoft clustering software.
Applications that use these other protocols will function identically to a
single-server environment. Users can still use these protocols, but they will
connect directly to the individual servers and not to the virtual servers on the
cluster, just as in a single-server environment. If a failure occurs, any
connections using these protocols will not switch over. Since these protocols
cannot fail over to another server, avoid these protocols, if possible.
Designing the Compaq ProLiant Clusters HA/F100 and HA/F200 2-37