Re: Melting the Internet?

Andrew Lih (
Thu, 04 May 1995 12:24:14 -0400

In message <9505041513.AA07745@frisky.Franz.COM>, Sean Foderaro writes:
> My message was that the current version of CU-SeeMe is internet-unfriendly
> because it uses UDP and doesn't do adequate flow control, and
> that it unnecessarly retransmits parts of pictures that haven't
> changed. The folks at CU didn't agree or disagree with this statement, they
> just said that the next version (due out within a few weeks) won't
> have these problems.

"Using UDP" doesn't necessisarily mean "being unfriendly." If the
software did keep track of ACKs and knew of what 8x8 blocks made it
through, then it could minimize sending blocks through for refreshing
"damaged parts." However, you're talking about a lot of state
information. The responsibilities of Reflector would exceed
"reflection," and would probably need to be renamed "Scorekeeper."

> I agree that in theory using UDP is better, as real-time data has
> different properties than static data. Whether it is better in
> practice will depend on how well CU layers flow control and
> eliminates unneeded retransmission. Also there are many different ways of
> implementing the TCP/IP protcol suite and often the only way to
> find out how well something performs is to experiment [see "TCP/IP
> Illustrated, Volume 1" by Stevens]. I've started some experiments
> and if anything interesting comes from it I'll publish the information
> here.

The "reliable connection" and "ordering" semantics of TCP make it
entirely inappropriate for packet video. You cannot drop packets --
your data stream is locked into end-to-end reliable delivery, at the
expense of timely arrival.

As for further information on the behavior of TCP, see the following
paper (and other related papers) from Ramon Caceres from AT&T BL for
an interesting look at the behavior of TCP. Although the paper
pertains to mobile operation, it importantly points out that TCP's
congestion window backs off with enough haste to put siginificant
delays into human-interaction applications.


> And one final reason for my message: I'd like to seen this
> mailing list used for some technical discussion.

That'd be great, continue on!


`''' Andrew "Fuz" Lih Columbia University
c @@ Mobile Computing Laboratory
- (PCMCIA: People Can't Memorize Computer Industry Acronyms)