To be redundant, the cluster interconnect must use two or more independent connections and data paths. An example of a redundant cluster interconnects is as follows:
Only one port is used for each Ethernet board. If more than one port of the same board is used as a cluster interconnect, these ports would be a common point of failure.
Connections to nodes must be independent.
No two cluster interconnects can share the same hub, switch, or routers.
The figure below shows a typical four-node cluster. In this diagram, there are two switches. Each of these switches has their own power connection on different circuits (when connecting two switches, it is necessary to use VLAN to detach each switch logically). If the switches were connected to the same power strip in a rack, for example, the power strip would be a single point of failure for the cluster interconnects.
Figure 3.2 Typical four-node cluster
