North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: audio/video again

  • From: Dick St.Peters
  • Date: Sun Aug 04 20:18:24 1996

> > Does nobody else even care that dialup users can request streams of
> > far more packets than can go down their dialup lines?
> 
> Is this a case of 14.4k users requesting 28.8k audio UDP streams or is the
> problem more widespread?

The case where I first ran into this was CU-SeeMe between a univerity
student, apparently on at least a T1, and a 14.4 dialup.  I don't have
a precise measure of the traffic, but it was around 30% of capacity of
what was then a 384 incoming feed.  Not much of it fit down the
dialup, but it all went through Pennsauken.

> It might be time for something like a "Bandwidth Conservation Society"

I think "conservation" is not the right concept.  The bandwidth is
there to be used, and people ought not to be discouraged from using
it.  I don't think flooding a destination with traffic it can't
receive counts as "use".

Sending no more than will actually arrive is only a minimal
requirement, not a sufficient standard.  Having observed streaming UDP
push TCP completely aside on a 56k, I imagine much the same thing
happens on any congested link: TCP will yield bandwidth until UDP has
all it needs to get the entire UDP flow through.

Sending no more that it's "fair" even to try to get through the most
congested portion of a route would be a much more stringent standard,
one that TCP meets according to my admittedly not very deep
understanding of how things actually work.

Of course, if simplistically applied, such a standard would completely
kill any realtime-ish applications on today's Internet, and I don't
want to do that.  Maybe an answer could lie in the definition of
"fair" involving a longer time scale - like it's ok to use UDP to
bulldoze your "realtime" flow though a congested link when competing
with lots of TCP flows if you don't expand to use more than your "long
term share" when not competing with many TCP flows (where the
assumption is that what's important for most TCP flows is the total
duration, not intra-flow speed variations).

Now that I've completely wandered off the deep end, I'd really like to
have some people who know what they're talking about say something.
All I really know for sure is that a lot of audio/video traffic
crosses the net only to not be able to go down the last phone line.  I
think it's likely to get worse real fast:
        28.8  -->  14.4
        ISDN  -->  28.8, 14.4

--
Dick St.Peters,       Gatekeeper, Pearly Gateway, Ballston Spa, NY
[email protected]     Owner, NetHeaven 518-885-1295/800-910-6671
Internet for Albany/Saratoga, Glens Falls, North Creek, & Lake Placid
	  First Internet service based in the 518 area code
- - - - - - - - - - - - - - - - -