On Fri, Dec 13, 2013 at 8:27 AM, Dan Stromberg <drsali...@gmail.com> wrote: > On Thu, Dec 12, 2013 at 6:16 AM, Grant Edwards <invalid@invalid.invalid> > wrote: > >>> Sockets reserve the right to split one socket.send() into multiple >>> socket.recv()'s on the other end of the communication, or to aggregate >>> multiple socket.send()'s into a single socket.recv() - pretty much any way >>> the relevant IP stacks and communications equipment feel like for the sake >>> of performance or reliability. >> >> Just to be pedantic: _TCP_ sockets reserver that right. UDP sockets >> do not, and do in fact guarantee that each message is discrete. [It >> appears that the OP is undoubtedly using TCP sockets.] > > I haven't done a lot of UDP, but are you pretty sure UDP can't at > least fragment large packets? What's a router or switch to do if the > Path MTU isn't large enough for an original packet? > > http://www.gamedev.net/topic/343577-fragmented-udp-packets/
I'm no expert on this (mostly I do TCP, or UDP with fairly small packets), but the packet should be reassembled at the far end. When your application comes to receive it, it'll receive the entire UDP packet as a whole. UDP fragmentation has several problems. First, if any fragment is lost, it won't be retransmitted (as TCP will), so the whole datagram is lost. And secondly, if you stream data across the network in a series of packets just a little too large to fit, each one will get split in two and you'll end up with twice as many packets going out, ergo abysmal performance. With TCP, there's the chance that the sender and receiver can between them figure out what packet size to use (cf path MTU discovery), but that won't happen with UDP unless the application consciously does it. So it's something to be cautious of in terms of performance, but if you want to send large UDP packets because they make sense, just go ahead and do it. Now, if you want reliability AND datagrams, it's a lot easier to add boundaries to a TCP stream (sentinel or length prefixes) than to add reliability to UDP... ChrisA -- https://mail.python.org/mailman/listinfo/python-list
