Networks guru Andrew Odlyzko recently estimated that a typical mobile user consumes 20MB of data a month for voice service, but that T-Mobile Netherlands reports their iPhone users consuming 640MB of data a month; so upgrading everyone to the Jesus Phone would increase the demand for IP bandwidth on cellular networks by a factor of 30.
It had in the past been estimated that major European cellular operators might be able to provide 500MB/user/month without another wave of network upgrades; if this calculation is at all typical, it looks like there is a substantial risk of an ”iPlayer event” hitting cellular in the near future. Recap: when the BBC placed vast amounts of its content on the Internet through its iPlayer service, DSL traffic in the UK spiked; or rather, it didn’t spike, the trend shifted permanently upwards.
That, of course, is much more worrying; because the marginal costs are set by the capacity needed to handle the peaks, a rise in average traffic means a boost to costs multiplied by the peak/mean ratio. An aggravating factor is the pricing structure for BT Wholesale backhaul service – the commits are 155Mbits/s, so if the new peak demand just exceeded your existing commit, you needed to buy a whole 155Mbits/s pipe. The impact on the UK unbundling/bitstream ISPs has been serious and the sector remains in a critical condition.
Traditionally, a mobile base station was provisioned with 2 E-1 leased lines, 2×2 Mbit/s capacity. Multiplied by 4, that’s 9,676,800 Mbits in a month. Divide by 8 to convert to MB, 1,181GB/1.15TB a month. Which means that a typical cell site could support at the most 1,832 users’ activity, or quite a lot less when you consider the peak/mean issue – typical values are 4:1 for GSM voice (458 users), but as high as 50:1 for IP (36!). Clearly, those operators who have had the foresight to pull fibre to the base stations and, especially, to acquire their own infrastructure will be at a major advantage.
The elements of traffic generation
The iPlayer event was an example of content push – what changed was the availability of a huge quantity of compelling content, which was also free. If Samsung’s recently announced video store takes off, that would be another example of content push. But this is far from the only driver of traffic generation, though. It is important to realise that the Internet video market is a tightly-coupled system. The total user experience is made up of content, of the user interface, of feedback and discovery mechanisms, of delivery over the network, and of the business model. All of them are very closely related – if the product is heavily DRM-restricted, prettying up the front end doesn’t help.
It is characteristic of a coupled system that the slowest-changing factor is the main constraint, but the fastest-changing factor is the driver of change. In this case, the slowest-changing factor is the infrastructure, and within that, the digs and poles of layer zero. Even the copper changes faster than that. The fastest-changing factor is the user interface, which can be changed at will. Sociability, discovery and the like, which require serious software development, are in the middle, with issues like BT Wholesale pricing some way below.
There was not much special about the iPhone technically; the first ones were 2G devices in a 3G world, and good luck to you trying to pull 640MB a month on GPRS alone. Is that even possible? Its integration with iTunes gave it access to content, but the cost issue meant that the bulk of the music on iPhones was probably downloaded over WLANs or sideloaded from a PC. But one thing that it did do very well was the user interface; Apple exploited its historic speciality in industrial design and GUI design to the limit. Typically, a lot of geeks and engineers scoffed at the gadget as an overdesigned bauble for big-kid hipsters; fools that we were.
But the core insight of the iPhone designers was to design for the Web and for rich media, probably helped by not having a telephony background. Therefore, they chose to cover as much of the form factor with a high quality screen as possible, and worked from there. They also made some advances in the GUI (zooming, gesture recognition), but the much talked about browser was less sensational. (Like all versions of Safari, it is based on the open-source WebKit engine that also makes the Nokia browser and Konqueror work.)
So we’re now beginning to see that changing the user interface can radically impact the engineering and economics of the network; and because it is a fast-changing element, it can do so faster than the network layer can react.
From receiving to sending
The Internet is a copying machine, they say; more to the point, it is usually a one-to-many medium that is experienced as a many-to-one medium. I draw content from many different sources according to the stuff I like; but each source is broadcasting itself to many readers. As a rule, people read more than they write, even if P2P distribution blurs this. One criticism of the iPhone is that it’s optimised for passive consumption of content; some users report their uplink/downlink ratio changing dramatically on changing to the iPhone.
Looking at another online-video sensation which hammers the ISP economy, YouTube, it’s quite clear that another driver of traffic is improved content ingestion. As whatever you place on the Web will be written relatively few times and read many times, there is a multiplier effect to anything that makes it easier to create or at least to distribute content.
YouTube’s innovation was three-fold; it made it dramatically simpler to upload video to the Internet, and it made it dramatically simpler to popularise it once it was there, through the embedding process and through its social functions. This latter feature meant there was much more of an incentive to upload stuff in the first place, because it was more likely to get viewed.
Better user interfaces and social mechanisms for content creation, then, are potentially major drivers of change in your cost model. They can change very quickly; and their impact is multiplied. Already, I can uplink photos to Flickr faster from my Nokia E71 than from my DSL link; granted, this is because of the UK’s lamentable infrastructure, but it shows some idea of the possibilities. Perhaps that Samsung device with the mini-decks might be less silly than we thought?
Faster adaptation: considered helpful
As we were wondering what would happen to the cellular networks’ backhaul bills, and contemplating the wreck of the DSL unbundler/bitstream business model, we looked enviously across the Channel to Telco 2.0’s favourite ISP, Iliad. They have just announced another set of fantastic figures; their margins are 70%-80% where they have deployed fibre, and their agility in launching new services doesn’t need to be rehearsed again. They even built their own content-creation service, after all; no fear of the future there.
What makes the difference? Iliad has always been committed to investing in engineering and infrastructure, giving it the agility to match the speed of change the application layer can achieve. It’s been determined to realise the OPEX and unbundling/wholesale savings from fibre deployment; and Iliad’s results have demonstrated that they are real and they are enough to fund deployment.
There is a crucial element, however, in their success; in France, access to duct and pole infrastructure is a regulated product, and major cities are more than keen on selling access to their own physical infrastructures – the sewers of Paris are the classic example. If you want to fix the ISP business model, fixing layer zero is the place to start, before the next fast-changing application knocks us back into the ditch.
- The ISP/telco market is a closely coupled system: An analysis in terms of differential rates of change shows that rapidly changing applications and user interfaces can have seismic impact on slowly changing network operator business models
- The benefits of fibre are real: Iliad is showing that fibre deployment isn’t just nice to have, it’s saving the ISP business model
- Open access to infrastructure is vital: There is no contradiction between applications/VAS and layer zero – instead they go together. If you want fantastic new apps, pick up a shovel.