[6bone] Problems with big packets!?

[Hugh LaMaster]

On Fri, 4 Oct 2002, Arien Vijn wrote:

> Alvero,
> 
> Your system undoubtedly has to fragment the data. After all the path 
> MTU size is most likely smaller than 5000 bytes.
> 
> I suggest you have a look at the packets (using tcpdump, ethereal or 
> some other fancy analyzer). You probably will see ICMPv6 Time Exceeded 
> messages, telling that the fragment reassembly time is exceeded.
> 
> Kind regards, Arien
> 
> On vrijdag, oktober 4, 2002, at 12:40 PM, Alvaro Vives wrote:

Several messages have suggested that packet sizes > 1500 bytes,
or even 1000 bytes, are unreasonable.  I certainly wouldn't
expect the existing 6bone to support packets > 1500 bytes,
but, packets of length 9000 over WANs are quite reasonable.

Existing ATM and POS links usually default to 4470, but, 
sometimes default to 9180 and certainly settable to 9180 
or even larger (e.g. Cisco GSR Engine 0 OC-3/12 POS cards
and Engine 2 OC-48 POS cards support MTU 17994,  
some GigE switches/ports support 9180, and some NICs 
support 9000.  People trying to do high-bandwidth applications
over WANs with large RTT's would be well-advised to set up the
path with 9180/9000 end-end.  

It would be unfortunate for high-bandwidth applications if only 
1500 byte packets end up being supported by IPv6.

--Hugh LaMaster