The penguin lvs daemon, or plvsd, runs on each node in an LVSM cluster.
It is in charge of maintaining all of the state in the cluster.
This consists of the ipvs routing tables, heartbeat's /etc/ha.d/haresources file, the config files for mon, as well as any ip aliases / tunnel interfaces on the realservers.
At startup, each plvsd reads the /etc/plvsd.conf on the local machine and then tries to get a consensus on what the proper config should be. Once all nodes have agreed on the configuration of the cluster, the cluster services are brought up.
Vips are managed through heartbeat, so when the cluster wants to bring up a vip, it writes a line to /etc/ha.d/haresources on each machine which tells heartbeat who the primary and secondary directors for that vip should be. Heartbeat then figures out which of the directors to bring the service up on. The actual vip ip is then brought up as an alias on the active director, and plvsd is notified. Upon notification, plvsd adds the correct rules to the ipvs routing tables and also starts mon to see which of the realservers for this vip are currently active.