"Communication between any two different L3 segments always requires a router." Communication between any two different L3 segments usually but not always requ

"Communication between any two different L3 segments always requires a router." Communication between any two different L3 segments usually but not always requires a router. If the gateway on a host is configured to be the ip address of the host itself then the host will issue arp requests for any ip addresses it is trying to reach. In this manner hosts on different L3 segments but the same L2 segment can directly communicate. Some DHCP servers including the Microsoft DHCP server could be configured to always assign hosts the same gateway address as host ip address. You don't mention L3 switches which in today's context makes sense because most L3 switches are really just L2 switches with built in routing capabilities but this wasn't always the case. Cabletron used to be a major ethernet switch vendor. Their L3 switching solution included a server connected to certain of their switches. This server communicated directly with the switch and was responsible for a variety of tasks regarding vlans. One of these was the Layer 3 switching function. Hosts connected to the switch were configured with the same gateway as their ip address. As mentioned above this caused them to make an arp request for any ip addresses not already in their arp table. The switch would forward any arp requests passing through it to the management server. This management server had its own table of L2 to L3 address mappings for all devices connected to the switch. If the requested address was in this table the server would respond to the ARP request with the correct MAC address. The host would then use this MAC address for the ip address and the switch could properly direct the ethernet frame to the correct port regardless of L2 segment. If the server couldn't resolve the L3 to L2 address mapping it would have the switch forward the arp request to ports which in the server software had been designated as router ports. The routers connected to these ports needed to have ProxyARP enabled. These routers would then respond to the ARP request and the host would send packets for the ip address to the router which had responded. This is a little simplified because there were additional rules which could be configured regarding VLANs and which could take advantage of this between each other. Also the server could connect to multiple servers at a time spanning this across a multiple switch deployment.