IP Unnumbered Explained

This topic is to discuss the following lesson:


Hhhm, so the routers don’t care if they are not on the same subnet?

tnx alot man

Hi Timothy,

With IP unnumbered they don’t, it’s kinda a weird concept but it works.


Thanks Rene,

That was Amazing …

learn something new everday. wow.

Rene, you are the man. Thanks!

Thanks Angel!

Hello Rene, Is this the only way 2 Router get adjagency ?
I want to mean…till now we learn that 2 router make up adj if they share the same subnet…so if the two router have an ip interface with ip addrress in that subnet…
Now I’m discovering that the ip could be “broken” …in the way that 2 router get adj also if they don’t share a subnet??? I’m fantasizing??? :slight_smile:

Hi Francesco,

It’s strange but that’s correct. In my example above it works with EIGRP, it also works with OSPF.

You have to use ip unnumbered on both sides though, otherwise you won’t get an adjacency.


you don’t need to use IP unnumbered on both ends. The one without IP unnumbered will still check for valid subnet. IP unnumbered just disables it.

Hi Rene,
My question is
If R1 Serial Interface is configured for ip unumbered and R2 Serial interface is not configured for ip unumbered but instead it is configured for some different subnet, will that work??

Hi Rupesh,

This unfortunately won’t work. The side that doesn’t have IP unnumbered will ignore any IP packets that are not on the subnet of the IP address that you configured on the interface. For example, if you try to run EIGRP like I did in the lesson then R1 will accept the EIGRP packets from R2 (since it has IP unnumbered) but R2 will show “not on common subnet” errors.


How does IP un-numbered work from an L2 perspective? Each interface on the serial will answer ARP?

Hello William

First of all, the example used by Rene includes the use of a serial interface, which is a very common use of the IP unnumbered feature. For a serial interface, ARP would not be a problem, because ARP is not used on serial interfaces. If it’s a point to point interface, the encapsulation from Layer 3 (IP) to Layer 2 (HDLC or PPP for example) doesn’t require a Layer 2 destination address. If you’re employing a point to multi-point serial connection such as Frame Relay, then reverse ARP is used to determine the IP address from the DLCI. In this case, the reverse ARP determines the IP address (indeed the borrowed address) from the DLCI.

Because the IP unnumbered feature can also be applied to Ethernet ports, your question is indeed a valid one. In this case, ARP would function as it does when sending an ARP request for any other IP address. The only restriction, according to Cisco, is that you cannot configure static ARP for this particular interface.

I hope this has been helpful!


Hi Rene,

for an example i am using this scenario

   Lo1--->Router 1---------------Router 2---->Lo1

In Router1 i am creating loopback1 ip address is and Router 1 connecting towards R2 that interface (serial )i configured as ip unnumbered.
In Router2 i am creating loopback 1 ip address is and Router 2 connecting towards R1 that interface (serial) i configured as ip Unnumberd.

Now R1 can able to ping R2 without using any routing Protocol?


Hello Gowtham

No this would not function, with nor without a routing protocol. The reason is because the IP addresses of the loopbacks are in the same subnet.

If you were to ping from Lo1 of R1 to, the router would see that this IP address is on the same subnet as the Lo1 interface, so it wouldn’t even look at the routing table. It would attempt to send the packet out of Lo1, and would fail. Regardless of whether you use unnumbered or not, or if you use a routing protocol or not, this would fail.

The borrowing of the IP addresses takes place only for the purpose of installing an entry in the routing table so that IP packets on serial interfaces can be processed.

Now if the two loopbacks were in a different subnet, say and, then one loopback could ping the other IF a routing protocol has been configured between the routers to share those prefixes, OR IF static routing has been configured.

I hope this has been helpful!


thank you Lagapides for your reply.

IP unnumbered supports only point to point interface. in case i want to configure on gig port it is multi point is there any possibilities to change into point to point in IOS (any syntax)

Hello Gowtham

Ethernet technology is inherently multipoint. You cannot convert an Ethernet port into a point to point connection by using a command on the interface, otherwise it would no longer be Ethernet. There is no benefit in doing so either. So there is no command that would allow the unnumbered feature to function on an Ethernet port.

Beyond the unnumbered feature, from a topology perspective, if you connect one router to another via Ethernet, you are really creating a point to point topology, since there are only two hosts on that link. But this does not change the way Ethernet functions, it is still multipoint, it’s just that in such a topology, there are only two hosts, so communication takes place as if it is point to point.

I hope this has been helpful!


Hi Team,

I am in the process of building my lab from the ground up.

I will be using spine and leaf topology.
To be specific, I will be using layer 3 spine and leaf architecture.
3 spines and 2 leafs 3750g and 3560x’s. Leafs will be conncted to EsXi host and other end-point devices(physical and virtualised devices firewall,pcs, cisco IDE etc)

I will be automating using Python(offbox ofcourse wish ios supported on-box :frowning: ) to streamline the process and save time.

I am having two issues, I will be using ECMP(load balance and utilise all links on each leaf) point to point routed ports fabric between the spine and leaf. My routing protocol of choice is OSPF.

My issue is for example the point to point link will be using however they will take up large address spaces unnecessarily(I will probably never need them but just dont like wasting and trying to emulate production i.e datacenter fabric).
I could use /32 for each routed spine/leaf p2p link but I feel it’s not appropriate. The other issue is when automating there will be alot of variable to change in my code, I am trying to make the code reusable, for example code for ACLs, Code for VLAN creation and subnet, Code for OSPF network fabric interface advertisement etc.

so A) address space wastage(one for management and one for P2P links for the fabric B) To many variables to edit when writing a reusable automation python code for the control plane of each device.

My solution was to utilise IP unnumbered. However that created subsequent issue.

The theory was to create a loopback interface on the ML spine and leaf switches and use the ip unnumbered for the p2p routed ports as well as use the loopbacks for device management. PLUS there will be less OSPF adjacencies correct?

Practically this created another problem. first of all, I know troubleshooting will be required when building the network from the ground up. I guess that’s the whole point of this project to become a better network engineer :smiley:

Subsequently how will i ping the point to point fabric egress and ingress interface to troubleshoot.

How can i ping these physical interfaces when they have no IP, the response will always comeback from the loopback interface which will always be up/up even if the physical interface is down?

Is there any solution to this?

I am enjoying this entire process really and it really provides one with an insight when planning and designing a network!.

As a matter of fact I don’t think it’s possible to be a good network engineer without having a access to a physical lab or even hybrid lab.

I have never in my entire life used a whiteboard, i am using one now!

I feel prepared for my CCNA already!

many thanks again guys!