[dcc2] UDP Nat2Nat Traversal

[Riley White]

On Apr 26, 2004, at 5:26 PM, Trent Lloyd wrote:

> Hi,
>
>> I think you're referring to actually transferring files using UDP?
> Yes :)
>
>> If
>> so, I think it might be a lot more hassle than it's worth.
> Well, possibly, I thought I'd suggest the idea as its usefull..
>
>> In addition
>> to handling ACKs, you also need to handle packet ordering;
>> just
>> because one UDP packet is sent before another doesn't mean the packets
>> are going to arrive on the other end in that order.
>
> *nod*
>
>> Basically we would
>> have to re-invent TCP (transmission control...), but luckily that has
>> already been taken care of :).
> yes, however this requires that one of the machines have a public IP
> address, and this is often not the case.
>
> I am also aware of the complexities of transfers over UDP, however in
> some cases its usefull, although, I guess, for general IRC clients it
> might be more of just a hassle over the potential gain.

I also think it could be useful, particularly after reading Peter 
Green's mention of the stun method. I would propose the following 
tokens.

ACK
Specifies that the receiving application should verify delivery with 
ACKs. Chunk size could be a the token value

ORD
Specify that the sending application must order the outgoing data 
packets, and that the receiving application must verify the order.

A DCC2 send using UDP could require these two tokens, although they 
would cause unnecessary redundancy in a TCP send. There could be a 
better way of doing this, but I thought I'd throw the idea out here.

--Riley