[Nanog] ATT VP: Internet to hit capacity by 2010
jgreco at ns.sol.net
Tue Apr 22 11:47:02 CDT 2008
> > > IP multicast does not help you when you have 1000 subscribers
> > > all pulling in 1000 unique streams.
> > Yes, that's potentially a problem. That doesn't mean that
> > multicast can not be leveraged to handle prerecorded
> > material, but it does suggest that you could really use a
> > TiVo-like device to make best use.
> You mean a computer? Like the one that runs file-sharing
Like the one that nobody really wants to watch large quantities of
television on? Especially now that it's pretty common to have large,
flat screen TV's, and watching TV even on a 24" monitor feels like a
throwback to the '80's?
How about the one that's shaped like a TiVo and has a built-in remote
control, sane operating software, can be readily purchased and set up
by a non-techie, and is known to work well?
I remember all the fuss about how people would be making phone calls
using VoIP and their computers. Yet most of the time, I see VoIP
consumers transforming VoIP to legacy POTS, or VoIP hardphones, or
stuff like that. I'm going to make a guess and take a stab and say
that people are going to prefer to keep their TV's somewhat more TV-
> Or Squid? Or an NNTP server?
Speaking as someone who's run the largest Squid and news server
deployments in this region, I think I can safely say - no.
It's certainly fine to note that both Squid and NNTP have elements that
deal with transferring large amounts of data, and that fundamentally
similar elements could play a role in the distribution model, but I see
no serious role for those at the set-top level.
> Is video so different from other content? Considering the
> volume of video that currently traverses P2P networks I really
> don't see that there is any need for an IP multicast solution
> except for news feeds and video conferencing.
Wow. Okay. I'll just say, then, that such a position seems a bit naive,
and I suspect that broadband networks are going to be crying about the
sheer stresses on their networks, when moderate numbers of people begin
to upload videos into their TiVo, which then share them with other TiVo's
owned by their friends around town, or across an ocean, while also
downloading a variety of shows from a dozen off-net sources, etc.
I really see the any-to-any situation as being somewhat hard on networks,
but if you believe that not to be the case, um, I'm listening, I guess.
> > What seems more likely is that we'll see an evolution of more
> > specialized offerings,
> Yes. The overall trend has been to increasingly split the market
> into smaller slivers with additional choices being added and older
> ones still available.
Yes, but that's still a broadcast model. We're talking about an evolution
(potentially _r_evolution) of technology where the broadcast model itself
> During the shift to digital broadcasting in
> the UK, we retained the free-to-air services with more channels
> than we had on analog. Satellite continued to grow in diversity and
> now there is even a Freesat service coming online. Cable TV is still
> there although now it is usually bundled with broadband Internet as
> well as telephone service. You can access the Internet over your mobile
> phone using GPRS, or 3G and wifi is spreading slowly but surely.
> But one thing that does not change is the number of hours in the day.
> Every service competes for scarce attention spans,
Yes. However, some things that do change:
1) Broadband speeds continue to increase, making it possible for more
content to be transferred
2) Hard drives continue to grow, and the ability to store more, combined
with higher bit rates (HD, less artifact, whatever) means that more
bits can be transferred to fill the same amount of time
3) Devices such as TiVo are capable of downloading large amounts of material
on a speculative basis, even on days where #hrs-tv-watched == 0. I
suspect that this effect may be a bit worse as more diversity appears,
because instead of hitting stop during a 30-second YouTube clip, you're
now hitting delete 15 seconds into a 30-minute InterneTiVo'd show. I
bet I can clear out a few hours worth of not-that-great programming in
> and a more-or-less
> fixed portion of people's disposable income. Based on this, I don't
> expect to see any really huge changes.
That's fair enough. That's optimistic (from a network operator's point
of view.) I'm afraid that such changes will happen, however.
> > That may allow some "less popular" channels to come into
> > being.
> YouTube et al.
The problem with that is that there's money to be had, and if you let
YouTube host your video, it's YouTube getting the juicy ad money. An
essential quality of the Internet is the ability to eliminate the
middleman, so even if YouTube has invented itself as a new middleman,
that's primarily because it is kind of a new thing, and we do not yet
have ways for the average user to easily serve video clips a different
way. That will almost certainly change.
> > I happen to like holding up SciFi as an example,
> > because their current operations are significantly different
> > than originally conceived, and they're now producing
> > significant quantities of their own original material.
> The cost to film and to edit video content has dropped
> dramatically over the past decade. The SciFi channel is the
> tip of a very big iceberg. And what about immigrants? Even
> 50 years ago, immigrants to the USA joined a bigger melting
> pot culture and integrated slowly but surely. Nowadays,
> they have cheap phonecalls back home, the same Internet content
> as the folks back home, and P2P to get the TV shows and movies
> that people are watching back home. How is any US channel-based
> system going to handle that diversity and variety?
Well, that's the point I'm making. It isn't, and we're going to see
SOMEONE look at this wonderful Internet thingy and see in it a way to
"solve" this problem, which is going to turn into an operational
nightmare as traffic loads increase, and a larger percentage of users
start to either try to use the bandwidth they're being "sold," or
actually demand it.
> > There are almost certainly enough fans out
> > there that you'd see a small surge in viewership if the
> > material was more readily accessible (read that as:
> > automatically downloaded to your TiVo).
> Is that so different from P2P video? In any case, the Tivo model
> is limited to the small amount of content, all commercial, that
> they can classify so that Tivo downloads the right stuff. P2P
> allows you to do the classification, but it is still automatically
> downloaded while you sleep.
I guess I'm saying that I would not expect this to remain this way
indefinitely. To be clear, I don't necessarily mean the current
TiVo device or company, I'm referring to a TiVo-like device that is
your personal video assistant. I'd like to think that the folks over
at TiVo be the one to leverage this sort of thing, but that's about
it. This could come from anywhere. Slingbox comes to mind as one
> > I'm envisioning a scenario where we may find that there are a
> > few tens of thousands of PTA meetings each being uploaded
> > routinely onto the home PC's of whoever recorded the local
> > meeting, and then made available to the small number of
> > interested parties who might then watch, where (0<N<20).
> Any reason why YouTube can't do this today?
Primarily because I'm looking towards the future, and there are many
situations where YouTube isn't going to be the answer.
For example, consider the PTA meeting: I'm not sure if YouTube is going
to want to be dealing with maybe 10,000 videos that are each an hour or
two long which are watched by maybe a handful of people, at however
frequently your local PTA meetings get held. Becuase there's a lot of
PTA's. And the meetings can be long. Further, it's a perfect situation
where you're likely to be able to keep a portion of the traffic on-net
through geolocality effects.
Of course, I'm assuming some technology exists, possibly in the upcoming
fictional Microsoft Whoopta OS, that makes local publication and serving
of video easy to do. If there's a demand, we will probably see it.
> Remember the human
> element. People don't necessarily study the field of possibilities
> and them make the optimal choice.
That's the argument to discuss this now rather than later.
> Usually, they just pick what is
> familiar as long as it is good enough. Click onto a YouTube video,
> then click the pause button, then go cook supper. After you eat,
> go back and press the play button. To the end user, this is much
> the same experience as P2P, or programming a PVR to record an
> interesting program that broadcasts at an awkward time.
I would say that it is very much NOT the same experience as programming a
PVR. I watch exceedingly little video on the computer, for example. I
simply prefer the TV. And if more than one person's going to watch, it
*has* to be on the TV (at least here).
> > I don't pretend to have the answers
> > to this, but I do feel reasonably certain that the success of
> > YouTube is not a fluke, and that we're going to see more, not
> > less, of this sort of thing.
> > > As far as I am concerned the killer application for IP multicast is
> > > *NOT* video, it's market data feeds from NYSE, NASDAQ, CBOT, etc.
> > You can go compare the relative successes of Yahoo! Finance
> > and YouTube.
> Actually, Yahoo! Finance is only one single subscriber to these market
> data feeds. My company happens to run an IP network supporting global
> multicast, which delivers the above market data feeds, and many others,
> to over 10,000 customers in over 50 countries. Market data feeds are not
> a mass market consumer product but they are a realtime firehose of data
> that people want to receive right now and not a microsecond later. It is
> not unusual for our sales team to receive RFPs that specify latency
> that are faster than the speed of light. The point is that IP multicast
> is probably the only way to deliver this data because we cannot afford
> additional latency to send packets into a server and back again. I.e. a
> type of solution won't work.
> It's not only nice to multicast this data, it is mission critical.
> People are risking millions of dollars every hour based on the data in
> feeds. The way it usually works (pioneered by NYSE I believe) is that
> they send two copies of every packet through two separate multicast
> trees. If there is too much time differential between the arrival of the
> packets then the service puts up a warning flag so that the traders know
> their data
> is stale. Add a few more milliseconds and it shuts down entirely because
> the data
> is now entirely useless. When latency is this important, those copies
> to multiple subscribers have to be copied in the packet-forwarding
> i.e. router supporting IP multicast.
> Of course consumer video doesn't have the same strict latency
> therefore my opinion that IP multicast is unneeded complexity. Use the
> tool for the job.
There are lots of things that multicast can be used for, and there's no
question that financial data could be useful that way. However, what I'm
saying is that this isn't particularly relevant on the public Internet in
a general way. The thing that's going to kill networks isn't the presence
or absence of the data you're talking about, because as a rule anybody who
needs data in the sort of fashion you're talking about is capable of buying
sufficient guaranteed network capacity to deal with it.
I could just as easily say that the killer application for IP multicast is
routing protocols such as OSPF, because that's probably just as relevant
(in a different way) as what you're talking about. But both are
What I'm concerned about are things that are going to cause major networks
to have difficulties. Given this discussion, this almost certainly
requires you to involve circuits where oversubscription is a key component
in the product strategy. That probably means residential broadband
connections, which are responsible for a huge share of the global Internet's
traffic. My uninformed guess would be that there are more of those
broadband connections than there are attachments to your global multicast n
etwork. Maybe even by an order of magnitude. :-) ;-)
Multicast may or may not be the solution to the problem at hand, but from
a distribution point of view, multicast and intelligent caching share some
qualities that are desirable. To write off multicast as being at least a
potential part of the solution, just because the application is less
critical than your financial transactions, may be premature.
I see a lot of value in having content only arrive on-net once, and
multicast could be a way to help that happen.
The real problem is that neither your financial transactions nor any
meaningful amount of video are able to transit multicast across random
parts of the public Internet, which is a bit of a sticking point.
Joe Greco - sol.net Network Services - Milwaukee, WI - http://www.sol.net
"We call it the 'one bite at the apple' rule. Give me one chance [and] then I
won't contact you again." - Direct Marketing Ass'n position on e-mail spam(CNN)
With 24 million small businesses in the US alone, that's way too many apples.
More information about the NANOG