[xcp] Embedding XCP signalling in IP header

[Marko Zec]

On Thursday 19 August 2004 00:41, Henderson, Thomas R wrote:
> Marko,
> Interesting proposal.  My thoughts are:
> i) it seems like it will be difficult to support the dynamic
> range and precision necessary with the fewer bits allotted.
> We experienced some problems with XCP convergence when
> fractions were being rounded. (we were using 16 bits
> for the feedback field, also with mantissa-exponent format).


Probably the most fundamental question so far!  Could you provide more details 
on your experiences?  Could the problem be linked to the congestion window 
increase, since the higher the packet/RTT sending rate, the more precise 
numeric representation for Delta Throughput and Reverse Feedback is required?  
Would it make sense to try dampening this issue by limiting the rate of 
sending packets containing DT / RF signaling to some reasonable value, say 
around 100 pkts/RTT?


> ii) given that a router has agreed to do XCP computations
> on your packets, wouldn't it be better to make its
> life easier by simplifying the arithmetic as much as
> possible?  I don't know whether heavy compression of the
> XCP fields is the right design trade to make.
>
> iii) shouldn't we be worrying about IPv6 anyway, by the
> time XCP is matured?


Of course we should. But we should also not ignore that IPv4 is here to stay, 
and that the lack of compatibility with current IPv4/TCP stacks equals the 
inability to interact with any of the existing systems / applications. If we 
could gain that compatibility, XCP would certainly see a much faster path for 
wide deployment and acceptance. That's what my proposal was all about...


> iv) Yongguang and I discussed previously the potential
> for reducing the frequency of packets that carry XCP-FWD,
> specifically to reduce the overhead consumed by XCP and
> to reduce the router operations, and I think it would
> be interesting to examine that angle from a sensitivity
> standpoint, although we did not look at it.


Yup, this would be well worth further investigation, regardless of the XCP 
header encoding scheme.

Cheers,

Marko