A server being multicast to does not distinguish between why a signal is not received, whether it is due the server failing or a network issue. If a signal is not received from a multicast server the server is marked as failed and the other servers stay in the cluster. If one of the four servers has network issue the other servers stay in the cluster.
Does a cluster not fail if multicast doesn't work in the weblogic environment ?
The servers with network issue should be marked as failed and not join the cluster. Otherwise how would it be known that why a signal is not being received from a server, whether it is due to network issue or the server has actually failed.
If one or two multicast messages lost, it doesn't really affect the cluster membership.
Internally Multicast Manager implements the heart beat mechanisam. It implements heartbeat retry protocal, if the multicast manager continuosly miss three heart beats then only server will be moved out of the cluster membership.
However, I would suggest customer to set the TTL values appropriately in their network also enable Message ordering enabled. If customer is in 11g, there are known issues make sure to suggest the respective patches. That will further help the customer to stable their environment.
Hope it helps.
If cluster spans several network paths or too many multicast packets are being lost Unicast is recommended. In Unicast each server has to inform only the Master not each other (as in Multicast).
For example, four servers WLS1, WLS2, WLS3, WLS4 with WLS1 as the Master. In Unicast each of the
WLS2, WLS3, WLS4 servers has to inform only the WLS1 server resulting in 3 network paths. In Multicast each of the WLS1, WLS2, WLS3, WLS4 has to inform each other resulting in 6 network paths.