UDP file transfers (Was: Re: [dcc2] Other comments)

[Bram Matthys (Syzop)]

Jeremy Iverson wrote:
> If we're talking about allowing UDP at the connection-negotiation
> level, I think that would be a good thing. Although it is unreliable,
> there are some applications where this doesn't matter (and hence, why
> UDP exists). 

I agree... Especially for video/audio etc.

myndzi wrote:
> I don't think UDP is needed but it is available so I use it in
> examples. It wouldn't be hard to number all the UDP packets, assemble
> them in the file out-of-order, and check a hash or something I
> suppose. In a file transfer getting the packets in order is NOT a
> concern. [..]

I really think *file transfers* over UDP is a bad idea.
Besides:
- I doubt it would be actually used
- I doubt it will be implemented by people

There are also some "real arguments" (which semi-result in the 2
points mentioned above).

When I think of UDP I think about streams/games/..: realtime data
communication where data may be lost and most of the time (but not all
the time) should not be retransmitted (think: games - player/object
positions, realtime video/sound, etc). In this case it's actually better
than TCP because useless retransmits would only delay the rest of the
data stream.

Now for file transfers where obviously the file would need to be
received 100% correct it makes no sense to use UDP.

- UDP packets can be send out-of-order, lost, etc, thus it would
   require _another_ protocol on top of UDP which basically redoes TCP.
   - This is a lot of work
   - You are semi-guaranteed to miss things
   - There's quite a big potential for (security) bugs
- It really makes more sense to use a stateful protocol here that has
   been in use and has proven to be functioning fine for over >20 years.

The only argument I heard pro- UDP filetransfers is that on some networks
UDP transfers would be possible but not TCP... I doubt this is often the
case. But furthermore, I don't believe this argument is so enormous more
valid than all the concerns I just mentioned. Note that it's not always
"the more options, the better".

Basically, based on the arguments I heard and thought of, I think
implementing UDP filetransfers is a waste of time and is guaranteed
to fail.

Regards,

	Syzop.

PS: Obviously I'm not writing this just to flame people to death,
     rather I try to share my views and experience here hoping to
     prevent certain (design) "mistakes". Then again, why would
     anyone think otherwise? ;)