[6bone] Problems with big packets!?

[Andy Furnell]

On Fri, Oct 04, 2002 at 04:28:13PM -0400, John Fraizer wrote:
> 
> 
> And how is it that you're seeing packets that are larger than the MTU of
> the interface?  If you run into that with a TCP packet, the bottleneck is
> going to send traffic back down the stream telling the sending station to
> reduce the size of the traffic.  You're not going to see that happen with
> an ICMP packet of 5000bytes.  It simply gets dropped.
> 

Different physical media have different MTU sizes. 1000 byte packets
aren't unusual, and can generally be transmitted across just about any
commonly used networking medium without fragmentation. Which is
something you seem to have forgotten. One of the major features of a
modern network is its ability to switch between different types of
physical network. A packet coming down the wire over ATM may be 4500
bytes. By your methodology when switched to an ethernet connection an
icmp packet would be returned telling the sender to adjust the packet
size it's sending. This is why we have the ability to fragment packets.

I think we're getting away from the point though. My point was that 1000
byte packets are a perfectly reasnoble size. Of most packets
transmittend I'd expect to see very few as small as 64 bytes (add 'in
the middle of a transmission' there if you want). Pinging with 64 byte
packets may be an accurate way to judge reachability to a host (and
latency to a certain extent), but if you're trying to simulate normal
internet activity packets closer to 1000 bytes are more likely to be
accurate.

A

-- 
Andy Furnell
andy@ipng.org.uk