NLB doesn't work when two nodes online

I am using Windows Network Load Balancing (according to this), the main purpose is building a High Availability (failover) for my RD Gateways, but I can't get the cluster worked properly. I have two nodes(both are VM, RD Gateway installed and only one NIC) in the NLB cluster:

Cluster Name: rdgw (virtual-ip: 192.168.0.18)

Host-1: rdgw-1 (192.168.0.21) (NLB installed, RD Gateway Server Farm joined, static IP, Converged)

Host-2: rdgw-2 (192.168.0.16) (NLB installed, RD Gateway Server Farm joined, static IP, Converged)

Custer Parameter -> Cluster Operation Mode: Multicast

Port Rule -> Filtering Mode: Single Host

Everything could be working when one of the two nodes is down!!

I already read this thread (NLB does not work when two nodes are present), but I don't think we have the same problem, cause I run all my nodes on OpenStack, so the Virtual-IP can map to the MAC addresses correctly.

I still have no idea how to solve my problem, below is what I have done for troubleshooting:

1. host-1 and host-2 can ping each other

2. From outside, ping 192.168.0.18 => DUP (two nodes respond to ICMP message, refer to this but doesn't work)

3. From outside, telnet 192.168.0.18 443 => Connection closed by foreign host

4. Check ipconfig /all in two nodes => Both nodes have two IP addresses, one is the original IP, another is Virtual-IP

IPv4 Address ......... 192.168.0.21(Prefered)
Subnet Mask .......... 255.255.0.0
IPv4 Address ......... 192.168.0.18(Prefered)
Subnet Mask .......... 255.255.0.0
Default Gateway ...... 192.168.0.1

I feel confused, I don't think this is a normal situation, this will cause IP Conflicts, am I right? To my understanding, Virtual-IP should only be configured on one of node at any given time.

I google some information about NLB heartbeats, it works with Layer 2, so there is no need to open specific ports for it.

I don't know what I am missing, maybe some miss configurations or heartbeats blocked by firewall ?

I would appreciate any help.