[OT] Handling multiple UDP clients through NAT
smorrey at gmail.com
Tue Feb 6 14:14:17 MST 2007
What about TCP limitations such as maximum connection counts etc.
How would I go about handling that or getting around it?
On 2/6/07, Bryan Sant <bryan.sant at gmail.com> wrote:
> On 2/4/07, Levi Pearson <levi at cold.org> wrote:
> > Solution: Use TCP. Keeping track of connections is what it was
> > designed for, and fortunately pretty much all network hardware is
> > designed to work with TCP connections, so you don't have to hack your
> > way around them.
> What Steve wants to do can be done with UDP. It's the layer 3 part of
> the TCP/IP stack that handles source/destination address:port
> information. UDP is built on IP (and TCP is built on UDP), so Steve
> can keep a unique list of users based on a source ip:port hash. So it
> is possible.
> However, I completely agree with Levi. Your application is inherently
> session based. TCP handles sessions for you (at least at the packet
> level). You will end up writing a crappy. bloated, version of TCP by
> the time you're done with this project anyway. Why not just use TCP?
> Game makers will often use UDP for LAN games because:
> A) LANs are reliable
> B) Even if there is packet loss, newer packets always have updated
> information (such as x,y,z location, etc), so there is no need for an
> aggressive, re-transmitting protocol like TCP.
> TCP prevents packet loss and is generally "A Good Thing" (tm) when
> sending data across the (unreliable) Internet -- hence HTTP being TCP
> based even though it is a stateless protocol.
> > But I strongly suggest you just use TCP.
> Me too.
> PLUG: http://plug.org, #utah on irc.freenode.net
> Unsubscribe: http://plug.org/mailman/options/plug
> Don't fear the penguin.
More information about the PLUG