Re: [dccp] CCID number in packet header?

[Eddie Kohler <kohler at cs.ucla.edu>]

Hi Ian!

This is very interesting, since the first proposed header I drew up  
for internal discussion in 2001 had the CCID in every packet!  So I  
have some sympathy for this.  We dropped it because of the space, and  
because I was eventually convinced that too many other parameters --  
features and stuff -- would be required before a middlebox could  
actually decode the connection sufficiently.

Eddie


On Aug 17, 2005, at 2:51 PM, Ian McDonald wrote:

> Hi there,
>
> I've been thinking a bit (dangerous thing I know) and wondered why we
> don't include the CCID number in every DCCP packet header.
>
> The main thing that I can think of NOT to do it is that it takes 1  
> extra
> byte of space (or 4 if you align on 4 byte boundaries and use new  
> space).
>
> The main reason why I would like to have the CCID in there is for  
> things
> such as network analysis tools (e.g. tcpdump) or for devices such as
> firewalls/routers. These can still work as is, but require much more
> work as you have to keep track of the feature negotiation and it is
> impossible if you start doing midstream in a traffic flow.
>
> What I therefore propose is that for packets with 48 bit sequence
> numbers that the CCID gets transmitted immediately after X as there  
> are
> 8 bits free here. For 24 bit sequence numbers we can leave it out as I
> think in reality this is less likely to be used and if really, really
> needed people can build the state.
>
> I realise that I am probably too late for this suggestion given the
> process of adopting RFCs etc, and there may also be reasons for what I
> have suggested being stupid so please feel free to make whatever
> comments you wish!
>
> Regards,
>
> Ian