Internet TV: Distribution and genres

There are various ways of distributing video in the internet. While some are suitable for all types of content, others can only be used for certain genres; and each has different costs for content providers, ISPs and consumers. Exploring these aspects will shed some light on what types of content can  be offered under the different business models of Internet TV.

This is a fairly technical post in that it uses jargon freely and assumes quite a bit of industry knowledge on the reader's part. It is also somewhat simiplified, as there are counterexamples to much of what I say; however, I think the distincitons I draw account for most of what is relevant today.

The various means of distributing video video in the internet fall into two groups:

• Streaming: A 'stream'  of video frames is sent over the Internet and played back at the consumer's premises with only a few seconds' delay.

• Download: A computer file with an entire unit of content (e.g. a TV programme or movie) is first prepared and placed on a server, and only then can a download commence.

This distinction is one between distribution mechanisms and not one of consumer practices; there is no reason why a consumer should need to be aware of what mechanism she is using.

Also, this distinction is not absolute and there are some hybrid models. However, barring some experimental set-ups, it is a fair description of the state of play today. For a more rigorous technical discussion see here.

In what follows I will discuss how well these means of distribution lend themselves to different types of content; what user experiences are associated to them; how the issue of rights affects their suitability; and how the costs are spread between content producers, transit providers, ISPs and consumers.

User experience

Streaming requires an uninterrupted connection between consumers and content publishers (or, in some cases, their intermediaries--see below). If bandwidth is not enough, viewing becomes impossible; if it is unreliable and drops frequently, video becomes pixelated or even freezes, resulting in an unsatisfactory experience.

Download supports two types of user experience: download-then-watch and download-as-you-watch.

• Under download-then-watch, a user first makes a decision about what content to watch, then starts a download, and only sits down to watch once the download has finished (which may be hours later).

• Download-as-you-watch allows users to start viewing while the download is taking place; with sufficient bandwidth (which, unlike for streaming, need not be stable), this yields an instant-gratification experience.

Under both models users watch content in the quality intended by its provider, without pixelating or freezing.

Genres

While streaming is suitable for all genres, download only works for recorded (i.e. non-live) content. This is because before a download can commence, a computer file with the required content needs to exist.

Streaming is thus the only alternative for genres such as breaking news and live sports.

The download-as-you-watch model is suitable for many televisual genres and practices, including "catch-up TV" where users watch a program that was recentily aired.

Download-then-watch is mainly suitable for "intentional" consumption where a user is clear about what she wants to watch, and to wait for it to download, and for "predicatable" viewing where a machine can start a download without an explicit request from the user (possibly using a recommendation engine, a user's viewing history, or her stated preferences).

A particular variant of download-then-watch is peer-to-peer distribution (P2P), which introduces the additional factor of latency: because peer-to-peer requires the desired content to be distributed to a sufficient number of users before a download can commence (and for popular content this number may need to be very large), it takes time for content to be easily available. This has the consequence that peer-to-peer is not suitable for content that needs to be consumed "fresh"--e.g. a news bulletin becomes stale after a few hours, P2P is not suitable for catching up on news.

Costs

One general observation can be made about costs without introducing further distinctions: download requires storage at the consumer's end, while streaming does not. While this is fairly irrelevant for PC-based viewing, it does mean that for a sofa-friendly experience, download requires more expensive set-top-boxes than streaming.

To explore the issue of costs in more detail, we need to take a closer look at our two means of distribution. Streaming and download each have two variants, each of which has a different cost structure.

Unicast streaming 

Currently used for nearly all web video, unicast streaming is the most expensive model for all parties. For content providers it requires maximum bandwidth as content needs to be sent in full to each consumer separetely. Further, a separate stream needs to be kept active for each consumer for the duration of the viewing experience, which requires sufficient serving infrastructure (i.e. servers). For transit providers and ISPs, unicast streaming also requires maximum bandwidth (while for the former this means revenue, for the latter it means costs).

Multicast streaming 

Although so far it has been a rarity in the public internet, there are projects underway to make multicast streaming a mass-market alternative. For content providers multicast is fairly inexpensive as only a handful of streams are required (thur requiring minimal bandwidth and infrastructure); for the same reason, for transit providers it represents minimal revenue. For ISPs multicast is relatively cheap as transit bandwidth costs are negligible; however, depending on the content's popularity, bandwidth usage within their networks can be considerable; and there are also costs associated to setting up and maintaining a multicast-capable infrastructure.

Setting up a multicast network accross the internet is a considerable organisational and commercial challenge. Further, once such a network is in place, the number of simultaneous streams it can carry is limited. As a result, multicast is suitable for live content of mass appeal such as sports events and TV channels, but not to niche content.

Standard download 

Standard download (i.e. FTP, HTTP or similar) imposes the same bandwidth requirements as unicast streaming on all parties. However, because (provided sufficient bandwidth) content can be downloaded faster than real-time (i.e. a two-hour movie could be downloaded in one hour), content producers can provide it with less serving infrastructure, and hence at a lower cost.

Peer-to-peer download 

Peer-to-peer download has negligible costs for content producers: not only is bandwidth kept to a minimum but also no streaming infrastructure is needed (indeed, under this model anyone can publish essentially for free). For transit providers and ISPs, peer to peer represents roughly the same bandwidth (and hence revenue for the former, and cost for the latter) as standard download and unicast streaming (this is somewhat different in the case of very large ISPs).

This cost must be borne by someone--namely ISPs from whose subscribers must be provide the content (thereby using bandwidth), and by consumers themselves who must provide the serving infrastructure (measured in disk space and CPU cycles). Although this arrangement has so far worked successfully with systems like BitTorrent, it is far from clear that this model will be workable in a mass-market, beyond-the-chasm scenario.It may well be that the only way of ensuring this is to force consumers to share their resources by refusing them access to content unless they share, or building this into dedicated equipment (i.e. set-top-boxes or software).