Which IP initiates health monitoring of nodes when F5 are in HA?

this is how it works. both nodes have to be able to determine for themselves if a poolmember is reachable. it makes sense to always use the non floating for this. else the health monitor IP would change with a failover.

As written by boneyard, the non-floating self-IPs are used for the monitoring. That´s the default behaviour and might be changed perhaps by using an external monitor.

That´s why your internal ACLs have to permit access to the real servers (aka poolmembers) for both non-floating self IPs and the IP-address being used for SNAT (i.e. the floating self IP with SNAT automap).

Be aware, that a monitor might even be fired through the out-of-band management interface (using the management IP). This will be the case, if there is a better route to the target and might be prevented by using route domains or the InTMM monitoring feature.