On Monday 17 April 2006 11:20 pm, Andrew Lentvorski wrote: > boblq wrote: > > On Monday 17 April 2006 06:04 pm, Stewart Stremler wrote: > >> TCP is hard to replace ... > > > > I agree. > > > >> you end up implementing something that looks > >> an awful lot like TCP. > > > > Not necessarily. One may blow off ack/nak and go with > > forward error correction ... and thus have a whole new > > thing with some very different attributes and issues. > > Surprisingly, you can't blow off ack/nack even with forward error > correction. The problem is that you *must* do rate/congestion control > otherwise your system goes pathological in a hurry. Rate/congestion > control requires that you ack at least some packets. At that point, you > might as well ack/nack lots of packets. > > TCP is really close to minimally optimal. Now, I just implement TCP > rather than trying to be "simpler". > > The only alternative I have seen is the Delay Tolerant Networking stuff. > Vint Cerf seems to think that TCP/IP won't work for interstellar stuff. > > http://www.dtnrg.org/wiki > > -a
I have been working with VOIP recently. As I was looking at RTP the old message above came drifting up from the back of my memory. RTP is UDP based and is _not_ reliable i.e. it is allowed to drop packets. It does not do ack/nak and hence does not worry about congestion control. Some background. http://www.voip-info.org/wiki-RTP http://www.cs.columbia.edu/~hgs/rtp/faq.html Given the growing volume of VOIP traffic I wonder if this is not a potential serious cause of congestion. Obviously another issue is QOS or rather its nonexistence. http://en.wikipedia.org/wiki/Quality_of_service I am curious what the real issues are here and what actual experience people have with them. There is quite a bit of research going on because of the importance of this problem. I found this http://www3.ietf.org/proceedings/06mar/slides/dccp-2.pdf There are other approaches including multi-tier networks. Comments, BobLQ -- KPLUG-List@kernel-panic.org http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
