Sunday, July 6, 2014

Are ISPs Responsible for Video Stalling?




Google's YouTube now is offering consumers reports about video streaming performance, as Netflix likewise displays messages that video stalling at the moment is caused by the Internet service provider.


To be sure, Internet service providers directly control contention ratios, capacity investments, network architectures and other network elements that enable and limit both bandwidth and latency.


On the other hand, other users also are, in a direct sense, the cause of congestion, as always is the case on shared networks.


And some apps impose more load on any network, video being the primary and common example, representing an order of magnitude or two orders of magnitude more bandwidth loading than other apps such as voice.


Compounding the problem, most streamed video entertainment is a zero-revenue driver of consumption, as are most apps, for the ISP. That is not to say apps should be sources of direct revenue for ISPs, but only to note that the business context for supplying more network resources is affected by that lack of revenue.


Any ISP will have a direct incentive to invest in additional capacity to support revenue-generating apps and services, and incentive to provide whatever level of network support is required to ensure good app performance.


Consumer Internet access is more problematic. On one hand, consumers expect affordable prices and virtually unlimited usage. On the other hand, suppliers have to balance expectations with a business case for more investment.


In other words, no ISP can afford, over the long term, to invest more in facilities and support than the customer is willing to spend to use the product, unless there are compensating revenue streams that can be used to augment the business case.


Voice services, texting, revenue-producing video entertainment or ancillary services are examples.


And there is no question but that widespread use of entertainment apps and services has dramatically different revenue implications for ISPs, compared to other apps.


How much bandwidth is required to earn those $43 revenue components? Almost too little to measure in the case of voice; gigabytes for Internet content consumption and possibly scores of gigabytes for video.


By some estimates, where voice might earn 35 cents per megabyte, revenue per Internet app might generate a few cents per megabyte. At one level, a network engineer might argue that such fine distinctions do not matter. The network has to be sized to handle the expected load.


McKinsey analysts have argued in the past that a 3G network costs about one U.S. cent per megabyte. The problem, in many developing markets, is that revenue could drop to as little as 0.2 cents to 0.4 cents per megabyte, for any mobile Internet usage.

That implies a strategic need to reduce mobile Internet costs to as little as 0.1 cent per megabyte, or an order of magnitude. Tellabs similarly has warned about revenues per bit dipping below cost per megabyte, leading to an "end of profit" for the mobile business.

The point is that ISP investment in higher-capacity networks does affect app quality. But so do prevailing business models, app bandwidth requirements, end user demand for video content and contention from other users.

Video at standard definition is one issue. High definition requires even more bandwidth. And "4K" video, requiring bandwidth four times that of HDTV, is coming. Few networks can be upgraded fast enough to cope with those sorts of capacity demands, to say nothing of changing business models to support continual capacity upgrades.

Yes, ISPs control their own investment levels. But they are not singularly responsible for quality of experience in the whole content ecosystem. Consumers, app providers, app technology requirements, display devices and communication networks and revenue relationships all play some part, even if it is the ISP that controls access bandwidth.

No comments:

Many Winners and Losers from Generative AI

Perhaps there is no contradiction between low historical total factor annual productivity gains and high expected generative artificial inte...