[6bone] Problems with big packets!?

[Hugh LaMaster]

On Fri, 4 Oct 2002, John Fraizer wrote:

> OK Hugh.  Just so the rest of the world doesn't get confused here:
> 
> END-TO-END is the key point when using MTUs >1500.
> 
> If you have a path that looks like this (MTUs):
> 
> [1500]-[4500]-[4500]-[17994]-[17994]-[4500]-[1500]
> 
> ...It doesn't matter what is in the middle, the max packet size
> (unfragmented) that is going to get through the chain is 1500.
> 
> With v6-in-v4 tunnels, 1500 won't even make it through.

True enough.  We're probably looking at MTU 1476 or somesuch
on these tunnels, regardless of whether or not the underlying
mtu on any particular link is 17994.  The tunnel is a virtual
link with a low MTU.

I just got concerned that some folks were generalizing this
phenomenon beyond the domain of tunnels and ethernet.  People
go to great lengths sometimes to get END-TO-END 9000 byte MTU
because it makes a huge difference in end-end performance
between hosts separated by a large RTT.  Actually, it makes
a big difference even if the hosts are locally connected and
on the same subnet, if the hosts are actually running user 
programs instead of just running network benchmarks ;-)