A preexisting network partition refers to a failure in communication channels that occurs while the systems are down and VCS cannot respond. When the systems are booted, VCS is vulnerable to network partitioning, regardless of the cause of the failure.
To protect your cluster from a preexisting network partition, VCS uses a seed. A seed is a function of GAB that determines whether or not all nodes have joined a cluster. For this determination, GAB requires that you declare the number of nodes in the cluster. Note that only seeded nodes can run VCS.
GAB automatically seeds nodes when:
When the last system starts and joins the cluster, the cluster seeds and starts VCS on all nodes. You can then bring down and restart nodes in any combination. Seeding remains in effect as long as at least one instance of VCS is running somewhere in the cluster.
You need to perform a manual seed to run VCS from a cold start (all systems down) when one or more systems of the cluster are unavailable. VCS does not start service groups on a system until it has a seed.