[e2e] bytes vs. packets

[Cannara]

Lloyd, how much MTU variability exists on typical paths?  I've yet to see
anything significant myself, and none in corporate nets, especially on VPNs.
I thought the conclusion was, when fragmentation statistics were requested
recently, that it was almost nil.

Alex

Lloyd Wood wrote:
> 
> On Sat, 9 Aug 2003, Dave Crocker wrote:
> 
> > I'd like to get a bit of discussion on this choice.  Not debate about
> > TCP, but clarification about the trade-off in the choice between
> > datagram-counting and byte-counting.
> >
> > The DCCP spec uses packets, not bytes, and it is being produced by
> > knowledgeable folk.  So it's worth being clear when each choice makes
> > sense.
> 
> In the presence of variable path mtu, counting datagrams doesn't give
> you a good idea of what's happening. In the presence of trivial,
> easily defeated congestion control algorithms, counting packets
> doesn't give you a good idea of what's happening. See RFC3465.
> 
> having wandered through
> http://www.ietf.org/internet-drafts/draft-ietf-dccp-spec-04.txt
> I imagine that there will be a number of similar, trivial, savage
> attacks on DCCP implementations based around Ack ratios and vectors.
>