[c-nsp] MTU on GigE

[Phillip Vandry]

On Sun, Nov 06, 2005 at 07:52:27PM -0500, Alex Rubenstein wrote:
> Question is -- does it work?
> 
> For instance, say you have a 7206, NPE-400, running 12.0S. It has a IO-FE. 
> So, int f0/0 can only have a 'mtu' of 1500, but you can set the 
> 'tag-switching mtu' to something much higher (say 9000).
> 
> Suppose further, that you are doing some AToM on this 7206, and the 
> interface you are AToM'ing is a DS3, with a MTU of 4470.
> 
> Will it work? My limited memory remembers that this did work in one 
> instance, but I can't say for sure.

I have not done this, but based on what you remember and also Dave
& Chad's experiences, it seems it can work. In a way, that is a
shame because it implies that the interface is perfectly capable of
passing such large packets and yet there is no way to make use of that
capability for IP: "ip mtu" is locked to 1500 for FE according to the
document I am reading, and even if it isn't, the physical interface
will reject the packets because the interface MTU is unchangeable
and is only slightly higher than 1500. This is where I read that tag
switched packets are exempt from the interface MTU. On the other hand,
I am skeptical of this document document because it also states that
IP packets that are accepted by the interface ("mtu") but exceed the
IP mtu ("ip mtu") may be fragmented on input, which I doubt.

> I thought the MTU issue on the FE cards of cisco was a physical one -- if 
> so, why does this work? What is the real reason cisco fe cards can't do 
> jumbo? non-technical?

There must be a physical limit somewhere:

xxxx(config-if)#ip mtu ?
  <68-1000000>  MTU (bytes)

surely I cannot successfully use 1MB packets!

-Phil