Does MTU on LOOPBACK matter?

RouterOS Forwarding Protocols
1

I have a CCR running OSPF, and run OSPF links at 1600 MTU. I noticed the Loopback MTU is 1500. Does an MTU have any meaning on the LOOPBACK interface?

19 R  sfp28-11-123main-fiber         ether             1600   1700       9570  48:A9:8A:18:83:C1
20 R  sfp28-12-123cal-radio          ether             1600   1700       9570  48:A9:8A:18:83:C2
21 R  LOOPBACK                       bridge            1500  65535             BE:14:C4:01:66:3B
22 R  bridge                         bridge            1600   1700             48:A9:8A:18:83:C3
23 R  gre0                           gre-tunnel        1476  65535
24 R  vlan363                        vlan              1600   1696             48:A9:8A:18:83:C3
25 R  vlan596                        vlan              1600   1696             48:A9:8A:18:83:C3

I ran a quick test and set the df-bit on pings to the router. At 1500 and 1600 I could still ping the LOOPBACK ip, (sanity check, at 1601 packet size, the packet did not make it there, as expected).

2

Note, we don’t run at 9000 MTU as wireless wires have a ~4k packet size limit, and we don’t (currently) have a use case for over 1600.

3

I don’t 1500 MTU matters on a OSPF loopback since it only used for routing messages. But there is also no harm in it matching the traffic.

4

Yes, it matters. I mean, wireless paths, technically, can do 9k MTU for layer 2 if the vendor supports, like some units from Ubiquiti.

But the point is, layer 2 MTU should always be MAXED out on ALL Devices, even if it’s different between them. Layer 3 MTU needs to be designed in a way that it ensures:

  1. PMTUD works correctly in ALL paths and directions (MPLS TE? LDP? OSPF? BGP? GRE? WireGuard? Bridge, VLAN, VPLS? Etc)
  2. L3 MTU should be 9k in the backbone on the physical ports, but you cap it wherever required on the layer 3 sub interface, such as LACP bonding to your upstream devices or the L3 VLAN interface towards another device etc

MTU design and training, is often something I don’t see very often in network training programmes and certs. MTU itself is a piece of cake, however, when your network has all types of devices with varying MTUs, proper planning is required. Don’t justify to yourself with this “1600” cap, do 9k MTU on the PHYs everywhere when possible to ensure future overhead room for anything.

My loopbacks are all 9k L3 MTU, most of my paths are 9k L3 MTU back-to-back, but some paths are 1500, some paths like wireless are 1500 on layer 3, but 1600 on layer 2 on the radios, but 9k L2 and L3 MTU on the transport ports on my PE router and P for example. Nothing breaks, because layer 3 MTU matches correctly between each termination L3, so PMTUD does it’s job. No packet fragmentation or VPLS Control word bullshit nor frame drops nor MTU mismatch.

You can follow the MTU section in this guide:
http://forum.mikrotik.com/t/how-to-edge-router-and-bng-optimization-for-isps/150007/1

And this:
https://stubarea51.net/2021/07/20/wisp-cheat-sheet-for-mpls-vpls-mtu/

Vlan/MPLS/VPLS issue V7 ?
5

@DarkNate is right here re maximize MTU.

I should have been clearer… I just meant it won’t break anything…

6

Isn’t there any disadvantage of using maximum L2/L3 MTU?

Higher lantecy? Memory/buffers usage?

7

MTU is a complex topic, that unfortunately is spread among many books and documentations.

The simplified version of it is:
Larger MTU = larger frames = larger IP datagram = larger pay load = fewer frames/packets required to complete the payload from start to end = less PPS required to transfer data from A to B = overall less utilisation of the path

I’m not sure why MTU deep-dive is not part of CCNA, JNCIA, MTCNA, Network+ or even in school/college.

I think MTU deep dive is covered only on CCDE and CCIE course work.