Clustering software is built on a distributed architecture where no one server is running by itself.  Either the application that you are using is running on multiple servers and the users connect to them all (Load Balencing) - or there is a dedicated backup server for each server that your users are connecting to that will kick in if the primary server fails (High Availability Cluster).  Both schemes use clustering software to manage the applications and servers that your users connect to.  Clustering software is redundant as well (as it should be ;-) ) so no single point of failure can cause the clustering software to fail.

This diagram gives an example of this.