Point-to-point (/31) addresses

CelticComms · August 11, 2013, 8:42pm

I’m not sure of the overall picture. Are these /32s public IPs you are assigning or /32s provided by an ISP at each location?

efaden · August 11, 2013, 8:50pm

Private addresses… just assigning them to the inside of a tunnel.

CelticComms · August 11, 2013, 9:30pm

If they are RFC1918 addresses why do you want to do anything other than use a /30? Conservation of non-RFC1918 addresses is what drives people into /31 & /32 land usually…

efaden · August 11, 2013, 9:36pm

I suppose mine was a theoretical experiment to learn about /32 addressing.

CelticComms · August 11, 2013, 9:56pm

If you add proxy-arp and use the eoip interface as the gateway(s) above then you will be able to ping those 10.x.x.x blocks both ways. However, whether that would be acceptable in a real world scenario might depend on a few other factors.

efaden · August 11, 2013, 10:54pm

Gotcha. Thanks for the info.

janisk · August 16, 2013, 1:33pm

it is easy to set up addresses using /32 addresses.

on router1 you set up

/ip address add address=10.10.10.10/32 interface=ether2 network=11.11.11.11

on router2 you set up mirror of that:

/ip address add address=11.11.11.11/32 interface=ether2 network=10.10.10.10

now you can manage routing or whatever you want to achieve via this. Key part is to set correct network address.

If you check PPPoE, it is working this way, so you can give out to customers real IP addresses but have a local address on the server (same for all tunnels)

Also, RouterOS do accept any other network mask there is(/30 /31), just you have to configure it properly. So if your main goal is to save addresses nothing is more efficient as /32 addressing.

StubArea51 · August 18, 2013, 3:23pm

Agreed…public Class A/B/C is the only reason I would mess with /32 and /31 routing. If you’re using 1918 space only, then do yourself a favor and use a /29. I very rarely provision /30 anymore in the private space between routers because you don’t have to redesign or configure if you want to use an IP in that subnet for troubleshooting or add a node.

pixitha · January 15, 2015, 9:32am

Thanks for posting on this topic guys, we use /31s all the time and i was really pulling my hair out trying to figure out why the MT wasn’t able to use them like other network gear. using the /32 on the MT solves it perfectly!

tomstorey · February 4, 2015, 10:12am

I want to suggest that actually this doesnt solve the problem. Its a work around.

To solve it properly, /31 should be supported.

StubArea51 · February 4, 2015, 5:22pm

I could go either way on that…it would be nice to have, but given all the development features requested of MikroTik , I’d rather see MPLS fast reroute or BGP ECMP before /31 addresses.

nicktc · July 24, 2016, 10:09pm

Sorry if I’m reviving a really old thread, however..

Using 10.0.0.A/32 to link router A-B-C together things that work great:

OSPF
MPLS
LDP
VPLS (point to point between A-B or B-C)

Things that do not work correctly

VPLS (between A-C no data RX on either router)

Adding an additional loopback address and using it as a transport address seems to “fix” the problem.

This is on 6.36

Anyone else seeing the same issue?

asy · February 27, 2018, 11:20am

Are there any plans to support RFC 3021 in RouterOS now ?

JimmyNyholm · February 27, 2018, 3:46pm

normis:

I would skip using an actual /31, and just use two /32s. Specify the remote address as the “network”, and you should be good to go.

This mechanism is more flexible than using /31s, as the addresses don’t need to be adjacent; and more efficient since you can re-use the same address for multiple links on a router.

In your particular case:

R1: /ip add add interface=ether1 address=10.10.10.2/32 network=10.10.10.3
R2: /ip add add interface=ether1 address=10.10.10.3/32 network=10.10.10.2

Will do the trick.

In a more complex hypothetical setup, with four routers connected to each other in a full-mesh of point-to-point ethernets:

R1: /ip add add interface=ether1-to-R2 address=10.10.10.1/32 network=10.10.10.2
R1: /ip add add interface=ether2-to-R3 address=10.10.10.1/32 network=10.10.10.3
R1: /ip add add interface=ether3-to-R4 address=10.10.10.1/32 network=10.10.10.4

R2: /ip add add interface=ether1-to-R1 address=10.10.10.2/32 network=10.10.10.1
R2: /ip add add interface=ether2-to-R3 address=10.10.10.2/32 network=10.10.10.3
R2: /ip add add interface=ether3-to-R4 address=10.10.10.2/32 network=10.10.10.4

R3: /ip add add interface=ether1-to-R1 address=10.10.10.3/32 network=10.10.10.1
R3: /ip add add interface=ether2-to-R2 address=10.10.10.3/32 network=10.10.10.2
R3: /ip add add interface=ether3-to-R4 address=10.10.10.3/32 network=10.10.10.4

R4: /ip add add interface=ether1-to-R1 address=10.10.10.4/32 network=10.10.10.1
R4: /ip add add interface=ether2-to-R2 address=10.10.10.4/32 network=10.10.10.2
R4: /ip add add interface=ether3-to-R3 address=10.10.10.4/32 network=10.10.10.3

Would fully interconnect them, using only 4 addresses. In this particular case, we get 3 times the address utilization efficiency of /31 addressing.

In an 8 router full-mesh the efficiency is even more spectacular, with only 8 addresses needed for /32 addressing, while a whopping 56 addresses would be needed for /31 addressing.

–Eric

This is the way I have tried this with routing protocols ospf and bgp and mpls it all works so my definite recommendation MT<->MT scenario is /32 and use the same IP for all local interfaces on one router. this eliminates the need for loopback interface as the router is the same IP to all interfaces. As MT doesn’t work well with /31’s the only way to interconnect MT<->OtherVendor is by /30’s

asy · February 27, 2018, 6:36pm

I read about MT-MT with /32.

As MT doesn’t work well with /31’s the only way to interconnect MT<->OtherVendor is by /30’s

But I need this. I linked MT(/32) - Linux(/31) but I can’t run OSPF over it (with Quagga). I tried nbma, broadcast
and point-to-point settings for OSPF but all unsuccessful. The best that I reach (with nbma, from log of Quagga):

OSPF: Packet 172.16.64.58 [Hello:RECV]: NetworkMask mismatch on eth2:172.16.64.59 (configured pref
ix length is 31, but hello packet indicates 32).

eth2/Linux: 172.16.64.59/31

RouterOS:
/ip address
add address=172.16.64.58 interface=ether1 network=172.16.64.59

So, does no way to run OSPF without /30 ?

mrz · February 28, 2018, 9:36am

Linux support /32 so switch to /32 and OSPF will work.

asy · February 28, 2018, 1:33pm

Linux support /32 so switch to /32

True. I switched “ip address add 172.16.64.59/31 …” to “ip address add 172.16.64.59/31 peer 172.16.64.59 …” and link continued to work. Ok.

and OSPF will work.

But this not. Quagga isn’t up neighbor with error (with nbma):

OSPF: interface eth2:172.16.64.59: ospf_read network address is not same [172.16
.64.58]

I attempted to switch ospf configuration to point-to-point. This gave some result: I can see neighbor in Quagga now:

172.16.64.58 1 Init/DROther 37.334s 172.16.64.58 eth2:172.16.64.59 0 0 0

but I doesn’ot see neighbor in RouterOS.

mrz · February 28, 2018, 1:38pm

show your quagga configuration

asy · February 28, 2018, 2:48pm

Sorry, I forgot about default firewall after reset configuration on MT. It works, thanks.

Linux:

ip address add 172.16.64.59/32 peer 172.16.64.58 dev eth2

Quagga:

!
  interface eth2
  ip ospf network point-to-point
! 
 router ospf
  ospf router-id 10.10.10.10
!
  network 172.16.64.58/31 area 0.0.0.112
!
  area 0.0.0.112 stub
!

MT (6.41.2)

/ip address
add address=172.16.64.58 interface=ether1 network=172.16.64.59
/routing ospf area
set [ find default=yes ] disabled=yes
add area-id=0.0.0.112 default-cost=1 inject-summary-lsas=yes name=stub-area1 type=stub
/routing ospf interface
add interface=ether1 network-type=point-to-point
/routing ospf network
add area=stub-area1 network=172.16.64.58/31

mrz · February 28, 2018, 3:16pm

change to
network 172.16.64.58/32 area 0.0.0.112