Let's consider an extreme example to explain how this works: a financial institution wants to place six PostgreSQL nodes, one witness server, and a backup system. This would clearly be a silly design, as shown in the following diagram:
This places every node in a single location that could lose its connection to the internet, lose power, or suffer some other catastrophe that would render the entire database stack unusable or even destroyed.
Now, let's apply the guidelines. First of all, we want to protect the backup; let's place that elsewhere, as seen in this following diagram:
Now, one PostgreSQL server and the backup are ...