[Discussion] MTU mismatch and impact of data-plane traffic

Mikael Abrahamsson swmike at swm.pp.se
Tue Oct 27 14:50:44 UTC 2015

On Tue, 27 Oct 2015, Mohamed Kamal wrote:

> Suppose you have the below network topology, where PE is connected to P1, P1 
> is connected to P2 and P2 is connected to GW, all through 1G links.
> [PE]-1500------------1500-[P1]-1600------------1600-[P2]-1500------------1600-[GW] 
> The numbers represent the MTU values configured in the following order; PE's 
> egress interface to P1, P1 ingress interface, P1 egress interface, P2 
> ingress, P2 egress and eventually GW ingress.
> Q1: What do you think would be the impact in terms of data-plane traffic 
> (HTTP/s browsing, Video streaming etc), traversing this network, in the 
> direction from the Internet and going to the PE router?
> My answer is:
> If there is a client running Win7 on a machine trying to access a web server 
> out there, the TCP MSS would be adjusted to around 1260-1460 bytes depending 
> on the Operating System's MTU value. Hypothetically, the first packet from 
> the web server destined to the client would be 1460-bytes and will reach the 
> ingress interface of the GW.

Where is the Win7 machine located in the above topology? If the end system 
uses 1500 as MTU, you won't notice the above misconfiguration because no 
packet will ever be bigger than 1500 (IP) so the mismatched MTU won't 
matter. Routers do not assemble packets (generally).

> Q2: However, what about larger MSS sizes? example; above 1500? and 
> larges chunks of payload from a connectionless protocols that don't 
> exchange MSS? UDP for example? or Google's QUIC (which is HTTP over 
> UDP)?

As soon as you try to send larger than 1500 byte packets in the above 
configuration you will star to drop packets (if MTU and MRU is the same). 
For instance, if the Internet is on the right and GW has 1600 to the 
outside, then you'll get a packet that is 1600 bytes that P2 will silently 
drop due to 1500 MRU.

Mikael Abrahamsson    email: swmike at swm.pp.se

More information about the NANOG mailing list