Re: The embarrassment of American broadband

[David Farber <dave () farber net>]

Begin forwarded message:

From: Rodney Van Meter <rdv () sfc wide ad jp>
Date: April 27, 2009 9:27:34 AM EDT
To: dave () farber net
Subject: Re: [IP] The embarrassment of American broadband


On Apr 27, 2009, at 7:26 PM, David Farber wrote:

> Begin forwarded message:
>
> From: Bill Stewart <bill.stewart () pobox com>
> Date: April 26, 2009 11:17:40 PM EDT
> To: dave () farber net, ip () v2 listbox com
> Subject: Re: The embarrassment of American broadband
>
>
> Yeah, Korea and Japan and even Sweden have really fast
> cheap service - but are they doing anything interesting
> with it besides TV, downloading movies from Pirate Bay,
> and occasionally looking at the produce at their
> neighborhood grocery store?  Where are the cool apps
> that drive new connectivity?  For dialup, it was email,
> and for low-end broadband, it was Napster and the web.

Okay, my family is (by design) a low-television family, but I'll bite:
what's so uninteresting about television as an application?

Most Japanese television is appallingly bad, making "Gilligan's
Island" look like "60 Minutes" and "In Living Color" look like
"Upstairs, Downstairs".  But, people do watch it -- although evidence
from ISPs suggests that more people are turning off the TV and turning
on the PC at night.

And here in Japan, all new TV sets are now required to include an
IP-capable connection (Ethernet, I believe).  When you buy a TV, the
salesman is supposed to talk you into a broadband connection for it,
if you don't already have one at home.  I'm a little fuzzy on what
application protocols they are expected to run, though.

During the recent World Baseball Classic, if we pressed the "Data"
button on the remote control, after a few seconds, the detailed box
score popped up on the screen.  It was cool, but the latency was
terrible.  (I know nothing about the protocols used to accomplish
this, but our TV is digital cable, and isn't yet plugged into the
network, so it's probably not IP, though who knows how the cable
provider actually does the uplink.)  We are, gradually, reaching the
interactive TV era that Nick Negroponte predicted so long ago.
See http://actvila.jp/ for example, a full on-demand service,
including for NHK's program archive.

How fat is an on-demand HD stream?  I've lost track these days (though
it's somewhere in the slides I use when lecturing).  How many people
are in your family?  Do they all watch the same TV at the same time?
It only takes two or three HD streams to justify a 50+ Mbps pipe.

What about videoconferencing?  H.264 at 768kbps is pretty good, but:
Anyone who has ever done an interactive conversation that way knows
that the conversation stops and starts, like talking to the Apollo
missions.  Usually the problem is just attributed to round-trip
latency, but that's at most a couple of hundred milliseconds.  The
real culprit is MPEG.  Using MPEG inherently results in high latency,
as it does inter-frame coding, requiring higher buffering; a GOP
(group of pictures) is typically something like 12 frames, or about
400msec.  , while the buffer->MPEGencode->xmit->buffer->MPEGdecode
times two total cycle can be well over a second.  Try running DVTS,
which is based on Motion JPEG (encoded a single frame at a time), in
parallel with H.264, and put them up on twin displays, and see the
difference in latency.  (That's a fun experiment we did between New
York and Tokyo during class last fall.)  See
http://www.dvts.jp/en/dvts.html

You think VC is uninteresting?  It is being touted as important in the
aging Japanese society, for telemedicine as well as reducing the
loneliness of elderly people unable to get out, and giving adult
children an extra way to stay in touch with their elderly parents.

Not radical enough for you?  GoogleEarth and Streetview thrive on high
bandwidth.  Joe Touch (former student of Dave Farber's, and one of my
best friends) is fond of saying, "Everyone complains about the speed
of light, but no one ever does anything about it."  That's a clever
line, but not really true.  Every cache, every prefetching operation
is founded on the hope that the data will be useful, allowing the
application to "beat" the speed of light.  With more bandwidth
available, you can prefetch more widely, more speculatively.

(DARPA/NSF managers, are you listening?  I think that further advances
in scalability require a more systematic approach to "latency-tolerant
computation", and one of the keys to that is "speculative
computation", which is a generalization of prefetching and optimistic
transactions.)

Not cool enough for you?  In Korea, they have professional networked
video game *leagues*, and matches are televised.  The top pros earn
what Tony Hawk does in the U.S.  Serious amateur players investigate
quite carefully the latency, bandwidth, and general reliability of
their network connections, and the Internet cafes there are filled
with players, despite the availability of bandwidth at home, because
the cafes are perceived to have lower latency to the critical servers.
(Okay, so that one's an ambiguous point about BW at home.)

Joe wrote a paper in the mid-90s that claimed that, most of the time,
if your application needs gigabit-class bandwidth, then it is
architected wrong.  You've put the data in the wrong place, or
designed the interactivity wrong.  At the time, he was largely right,
aside from the important class of Big Science apps; now, with things
like the above, I think the time has finally arrived.


All of that said, that's not even the most interesting thing that has
happened to the Internet in the last decade or so here in Japan.  That
honor has to go to the mobile phone-enabled Internet, i-mode and the
like.  Their models were simple enough, a decade ago, to be
implemented simply and quickly, and email (and now GPS-enabled maps
and the like).  Those models are now bumping up against some
limitations of their original architectures, but that's another
posting.


> I'm more interested in what you can do with the service
> than I am with absolute speeds - the big applications for
> speeds over ~1 Mbps are television and its competitors,
> which are intellectually uninteresting even though they
> may be important market drivers and may cost less than satellite.
> On the other hand, I really object to ISPs that won't let me
> run any kind of (non-spam) server on my home machines,

Hear, hear.

                --Rod





-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com