You're not giving much in the way of detail about what you're trying to accomplish (what service are you trying to make highly available, etc.) or how you're trying to accomplish it (I can assume you want to use native clustering, but you haven't specifically said so) and what hardware you have available to achieve those ends. Thus, advice would be limited to "generalized clustering best practices". To that end, want to make sure that you've eliminated any single points of failure (SPOFs) within the hardware you're rolling out. This means:

• You'll want to use redundant network interfaces on each host
• You 'll want to make sure your redundant network interfaces are spread across multiple NIC cards (i.e., don't use two ports on a multi-port card)
• You'll want to make sure your NICs are attached to redundant switches
• You'll want to make sure your switches are attached to redundant routers
• You'll want to either use redundant paths to your storage or make sure that each host is connected to replicated filestores 
• You'll want to ensure that you connect to your storage through redundant controller ports
• You'll want to make sure that those redundant controller ports are spread across multiple controller cards
• If connecting to SAN storage, you'll want to make sure that you're doing so through redundant paths (i.e., fault-isolated SAN switch-pairs)
• You'll want to make sure to connect to redundant storage processors on your storage array.
• etc.

Basically, look at the hardware you have available to you and identify any components that might fail and impact availability of a node or the entire cluster. Then, do what you can to mitigate the effects of such a component failure.