From Proxmox VE 4.0 and later, multicast is now required for proper cluster communication. In simple words, multicast delivers a single transmission to multiple server nodes in a network simultaneously, whereas unicast sends data packets to a single destination from a single source. The more nodes there are in a cluster, the more separate unicast packets need to be sent by it. Using multicast, this extra amount of traffic is vastly minimized. Due to the increase of packets in the network when using unicast, implementing it in a cluster with five or more nodes should be avoided. In order for multicast to work, the physical switch in the network must be multicast and IGMP snoop capable.

IGMP snooping is simply a process where the physical switch listens or snoops for an IGMP conversation between the nodes and the switch. This allows the switch to maintain a table or map to determine how and where to direct multicast requests. After enabling IGMP snoop, it takes a few hours for the switch to establish the table after gathering enough data for all multicast-enabled switch ports.

If it is not possible to use multicast at all in a Proxmox environment, then unicast is the only choice. To test whether multicast is functioning in the cluster, we can run the following command on all the Proxmox nodes:

# omping <remote_node_ip> <local_node_ip>   

If multicast is functioning fully, the output will show multicast responses:

If you only see an error message waiting for response msg, that means the command omping is only running on one node. Only by running it on multiple nodes simultaneously can we generate multicast traffic. Unsuccessful multicast responses will show packet loss for the node. The documentation of each physical switch should show whether the switch is multicast capable. However, nowadays, almost all smart and managed switches have the multicast feature. It is, however, disabled on all ports and must be enabled for proper Proxmox cluster communication.