Andrew Odlyzko, best known for his objective commentary on Internet traffic growth, spoke at the 2008 Gilder/Forbes Telecosm Conference. He helped clarify what really matters in the debate about Internet bandwidth growth. Consider this a survival guide in a time where “Internet traffic growth rates are slowing and Hype is accelerating.”
Andrew Odlyzko was one of the first to burst the misconception (I’m being kind with this word choice) that Internet traffic was doubling every 3 months. He has devoted great effort to gather hard data about Internet traffic growth, a valuable service when virtually all commercial and financial participants are biased towards predicting larger than actual growth. It is tough to believe Cisco’s prediction that traffic would grow 40% a year as it’s objectivity is hopelessly compromised by it’s market position. Does anyone really expect Cisco to predict Internet traffic below consensus opinion?
The MINTS project organized by Andrew is a step towards objectivity and compiles data from publicly available sources. It is considered the first and last objective measurement of internet traffic growth. A list of sources is here.
Here are some of the interesting points he made during his presentation (full .pdf here)
- “Should we worry about too much or too little Internet traffic?”
- “Internet traffic growth rates are slowing. Hype is accelerating.”
- The per-capita traffic intensity of Hong Kong is 6x the US level but is growing much slower (11% vs approx 50%).
- Telecom is the only industry that worries about it’s customers using too much product.
- Assuming traffic growth of 50%, carrier business is revenue neutral if costs drop 33%.
- Volume is not value. SMS messages consume almost no bandwidth but bill out at $1000/Mb
Based on Andrew’s data and our own work on Japan bandwidth growth (see “The Bandwidth Explosion Myth”) it appears that internet growth is maturing, and simply installing faster connections (like FTTH in Japan) won’t make a difference. The only thing that will kick-start growth again is new applications.
The application most often touted as the source of the next bandwidth is Video. CEO’s of equipment and chip companies highlight Video as a key factor driving growth. Pundits agree. Yet none can point to specific facts that quantify the claim. Cisco has shipped only 500 units of their much vaunted Telepresence video conference system (source). Ask a CEO to lay out, in detail, exactly how substantial equipment investment is being used to fund more bandwidth for video specifically. I can’t find one who can do so and would welcome the conversation.
The consensus opinion that video traffic driving bandwidth demand is unchallenged, but is remarkably similar to claims in 2000 that Internet traffic was doubling every 90 days. I would really like to see someone objective, like Andrew Odlyzko, look into the claim that video is driving the growth of the network.
The reality of video is it is highly cacheable. Take the top 1000 videos and store them at the edge of the network and you have eliminated a very large quantity of traffic. This appears to be what Zeuguma Systems is doing. Folks who cite “The Long Tail” and the abundance of niche content forget that The Long Tail isn’t very fat. There are only so many folks who want to watch oddball videos. Most people think they are special but in reality just want to watch “American Idol”.
The most important conclusion I drew from Andrew’s presentation and the panel that followed was the following: Traffic growth simply doesn’t matter. Period. What matters is revenue. Arguments about an ‘Exaflood’ or Video bandwidth explosion are not meaningful unless their deployment results in incremental revenue that carriers can use to fund new investment. And it is very difficult to see how Internet Video (not IPTV, that is different) puts more dollars into the pockets of carriers that they can hand over to Ciena, Alcatel, and Huawei.
I think that the game-changer in video will be user-generated content. Not a bunch of guys filming reality shows in their backyards, but the “send to grandma” home videos.
If I put a picture up on Flickr or a Picasa web album, and someone looks at it, they’re maybe consuming 5kb/s – 50kb/s (depending on how long they look at it).
If I put a video up on YouTube and someone looks at it, they’re consuming about 325kb/s – one to two orders of magnitude more.
But with a $1000 HD camcorder, I can generate HD video, and grandma doesn’t want to see crappy YouTube-quality videos, she wants to see every freckle on the grandkids’ faces. So an HD-quality video-sharing application will be consuming maybe 8-16 Mb/s – another one to two orders of magnitude. (Sure, AT&T will say that they can compress HD down to 6.5 Mb/s, but that’s with $100k professional-grade MPEG-4 encoders, not the encoders in a typical consumer-grade camcorder.)
It’s very easy to generate immense quantities of video that a very few people will want to watch. If a user can share an HD video with a few people just as easily as they can share a dozen pictures today, the bandwidth requirement goes up by three orders of magnitude.
from UBS/John Hodulik’s note today
IP bandwidth pricing continues to tumble
Cogent announced yesterday it had changed its pricing structure for IP transit to other service providers and introduced volume discounts for its “net-centric” customers. The company has long advertised its $10/Mbps standard pricing for bandwidth. Customers signing one-year ($9/Mbps) and two-year ($8/Mbps) contracts could avail of discounts off the sticker price. Last month, the company began offering its carrier customers a price of $7/Mbps when signing a 3-year contract. The more dramatic pricing change comes for “net-centric” (Content providers) customers. Mgmt began offering volume discounts on a sliding scale starting from commitments ranging from 100 MB/month with the lowest price of $4/Mbps for customers commiting to 10 GB connection and signing a 3-year contract. Management believes this pricing will stimulate growth of internet video.
Cogent has long been the price leader in the industry offering $10/Mbps for bandwidth compared to average industry pricing of $30/Mbps. CEO Dave Schaeffer suggests the company is taking significant market share as IP traffic on its backbone has grown at a 5-year CAGR of 140% compared to overall internet growth of 75% annually. However, last year growth on Cogent’s network slowed considerably to 75%. Mgmt suggested this is due to a slowdown in overall internet traffic growth to 57% in 2007. Cogent suggested internet traffic growth continues to slow as sequential traffic growth slowed by 700 bps in 1Q08 (from 14% to 7%). We note this contradicts Level 3’s assertion that overall IP traffic continues to grow in the 80-85% range.
Hi Andrew,
First off, great post. Thanks. Your last sentence opens the door to a great deal of speculation, if not debate, as well: “And it is very difficult to see how Internet Video (not IPTV, that is different) …”
Tell this to the bean counters whose interests are enhanced by claiming otherwise. Consider this softball question, for example: Is a telepresence session carried over an enterprise VPN tunnel considered Internet traffic? How does the answer to this question play with the part of the session that s carried over the enterprise’s dark-fiber network or dedicated private lines.
In counting the total number of Internet bits carried, do we now need to resort to performing bitrate-mile breakdown analysis in order to delineate what’s on-Net and what’s not? And whose “on-net” traffic is it, anyway? Add enterprise campus-area and metro-area to the dark fiber classification above, wherever they have replaced an ISP’s or carrier’s metro-edge facilities — which previously included up-and-down hierarchical routing while still “on” the Internet.
The point here being, without a Divine Authority to provide a detailed taxonomy of Internet complete with qualifying criteria, it’s difficult, if not impossible, to arrive at a consensus as to what is and what is not “Internet” in many respects.
Ever since the mid-nineties we’ve seen a migration of legacy traffic formats onto an “Internet” whose parameters have changed in lockstep with the meaning of the term “ISP”, which has changed, since the johnny-come-latelies have all tended to be the ‘incumbents’ telcos and cablecos of yore who are now enfranchised in a new domain by virtue of changing from ATM to quasi-all IP. And just as they’ve displaced the mom-and-pops’ role in Internet, so tpp have their traffic loads (PSTN-VoIP, IPTV, VPN, etc.) changed the texture and volume of Internet traffic to a very large degree.
These migrations have in large part been seen as “incremental growth” on the Internet, ignoring entirely the displacement effect I just mentioned above, which has actually taken place, i.e., the shift as opposed to absolute increase in total global bandwidth consumption.
Consider, for example, a service territory that is converted from HFC video to IPTV. All else being equal, given the conversion factors one must consider from “500 channel broadcast mode” GHz-miles to “single-channel switched mode” Mbps-miles when changing from analog to digital, has the conversion had the overall effect of increasing or decreasing traffic (no matter, whether Internet related, or otherwise)? Consider the various means by which backhaul is achieved by some LM providers prior to answering that last question, especially as relates to entities smaller than the Tier Ones who are also providing video in the last mile.
Much of this is viewed as “off-Internet”, to be sure, although it’s all IP (in name, at least) and ported around the metro and edge networks in various latter-day incarnations of Layer 2 transport encapsulating Layer 3, while using the same routers, the same switches, with much of the latter gear mounted in the same racks in the same meet-me locations.
I agree with the distinction you make when you note “that is different”. However, one might ask, “what is ‘not’ different, or exactly what, today, constitutes ‘Internet Traffic'”?
Lest I be misunderstood, I’m not disputing that overall traffic growth in all of its diverse forms is actually taking place. Instead, I’m simply pointing out the folly we often find ourselves engaged in when attempting to quantify it, never mind prove it.
As if to emphasize some of my earlier points, O posted the following to NANOG last year after a discussion with Andrew Odlyzko on the Cook Report list:
from: http://tinyurl.com/6e4lov
Subject: Re: Measurement data on transit traffic in IP routers?
Group: Nanog
From: Frank Coluccio
Date: 19 Feb 2007
>Your statement makes something of a presumption
>as to the architecture of a network. In many
>networks, edge aggregation devices do not
>participate in backbone routing, but simply
>pass the traffic they are aggregating into the core.
My first reaction, as well. However, I was reminded
by Andrew Odlyzko that the cable tv industry’s (MSOs’)
peering universe constitute a form of flattened ‘edge’,
if one were to consider the larger Internet’s core
against the MSO community, which makes for another
form of interesting analysis, since much of today’s
(especially more capacious) residential “broadband”
flows begin and end on MSOs’ networks, and sometimes
never touch the larger core, fwiw. And this opens the
door to other forms of “walled garden” environments,
including intranets, some providers’ CDNs, extranets,
and so on.
Trackbacks / Pingbacks